WEBB - The Physics of Three-Dimensional Radiation Therapy

Series in Medical Physics
THE PHYSICS OF
THREE-DIMENSIONAL
RADIATION THERAPY
Conformal Radiotherapy,
Radiosurgery and
Treatment Planning
Steve Webb
Professor of Radiological Physics,
Head, Joint Department of Physics,
Institute of Cancer Research and
Royal Marsden NHS Trust,
Sutton, Surrey, UK
Institute of Physics Publishing

Bristol and Philadelphia
Copyright © 1993 IOP Publishing Ltd.
0 IOP Publishing Ltd 1993
All rights reserved. No part of this publication may be reproduced, stored

in a retrieval system or transmitted in any form or by any means, electronic,
mechanical, photocopying, recording or otherwise, without the prior permission
of the publisher. Multiple copying is permitted in accordance with the terms of
licences issued by the Copyright Licensing Agency under the terms of its agreement
with the Committee of Vice-Chancellors and Principals.
British Librav Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.
ISBN 0-7503-0247-X
ISBN 0-7503-0254-2 Pbk
Libravy of Congress Cataloging-in-Publication Data are available
Reprinted with corrections 2001
The author has attempted to trace the copyright holder of all the figures and tables
reproduced in this publication and apologizes to copyright holders if permission
to publish in this form has not been obtained.
Cover picture depicts a multileaf collimator being used to shape an x-ray field
during radiotherapy. The collimator has many pairs of small lead jaws that generate
a field matched to the view an observer would have of the tumour from the source
position.
Series Editors:
C G Orton, Karmanos Cancer Institute and Wayne State University, Detroit, USA
J A E Spaan, University of Amsterdam, The Netherlands
J G Webster, University of Wisconsin-Madison, USA
Published by Institute of Physics Publishing, wholly owned by The Institute of

Physics, London
Institute of Physics Publishing, Dirac House, Temple Back, Bristol BS1 6BE, UK
US Office: Institute of Physics Publishing, The Public Ledger Building, Suite
1035, 150 South Independence Mall West, Philadelphia, PA 19106, USA
Typeset in ETfl using the IOP Bookmaker Macros
Printed in Great Britain at the University Press, Cambridge

CONTENTS
PREFACE ix
...
ACKNOWLEDGEMENTS Xlll
1 THREE-DIMENSIONAL RADIATION-THERAPY TREATMENT

PLANNING 1
1.1 Conformal radiotherapy treatment planning 1
1.2 Registration of two image datasets for 3D treatment planning 38
1.3 Summary and the NCI study of 3D radiation therapy planning 54
References 55
2 TREATMENT PLAN OPTIMIZATION 65

2.1 General considerations 65
2.2 The impossibility of true inverse computed tomography 67
2.3 The case of a circularly-symmetric dose distribution 72
2.4 Primitive blocked rotation therapy 77
2.5 Methods for 2D and 3D optimization 87
2.6 Summary 124
2.7 Appendix 2A. A historical note on the origins of rotation therapy 125
References 127
3 STEREOTACTIC RADIOSURGERY AND RADIOTHERAPY 135

3.1 Introduction 135
3.2 Radiosurgery and stereotactic radiotherapy 144
3.3 Stereotactic interstitial implant therapy 163
3.4 Summary 164
References 165
4 THE PHYSICS OF PROTON RADIOTHERAPY 172

4.1 Introduction: elementary physics of proton beams 172
4.2 Proton-therapy facilities 179
4.3 Range modulation and production of large-area beams 186
4.4 Proton treatment planning 198
4.5 Summary 212
V
vi Contents
References 213
5 CONFORMAL RADIOTHERAPY WITH A MULTILEAF

COLLIMATOR 218
5.2 Modem developments 223
5.3 Brahme's theory of orientation 233
5.4 Optimized beam profiles 235
5.5 How the leaf positions may be determined 237
5.6 Systems commercially available in 1992 238
5.7 Multileaf collimators for proton therapy 24 1
5.8 Summary 24 1
References 24 1
6 MEGAVOLTAGE PORTAL IMAGING 246

6.1 The need for high-quality portal imaging 246
6.2 Fluoroscopic detectors 247
6.3 Scanning linear array of diodes and crystal detectors 254
6.4 Matrix ionization-chamber detectors 259
6.5 Improvements to portal imaging with film as detector 264
6.6 Imaging with a photostimulable phosphor plate (Fuji system) 27 1
6.7 Portal imaging by reconstructing from projections 273
6.8 Portal imaging by xeroradiography 274
6.9 Solid-state imagers 275
6.10 Diagnostic imaging on a linear accelerator 276
6.11 Theoretical considerations of dose and image signal-to-noise
ratio 277
6.12 Portal dose images; transit dosimetry 278
6.13 Megavoltage computed tomography 280
6.14 Summary 28 1
References 283
7 TREATMENT MACHINE FEATURES FOR CONFORMAL

THERAPY 290
7.1 The earliest treatment machine for conformal therapy with a
'37Cs source 290
7.2 Tracking units 29 1
7.3 A tracking linac with multileaf collimator and CT combination 292
7.4 The universal wedge for the linear accelerator 293
7.5 The dynamic wedge for the linear accelerator 295
7.6 Wedges with the multileaf collimator on a linear accelerator 297
7.7 Linear accelerators with independent collimators 297
7.8 Two-dimensional tissue compensators 299
7.9 Summary 302

Contents vii
References 303
8 IMAGING FOR CONFORMAL RADIOTHERAPY PLANNING 306

8.2 Principles of imaging by computed tomography 307
8.3 X-ray computed tomography 311
8.4 Magnetic resonance imaging 313
8.5 Ultrasound imaging 3 16
8.6 Single-photon emission computed tomography-SPECT 3 17
8.7 Positron emission tomography-PET 3 17
8.8 CT imaging on a radiotherapy simulator 318
8.9 Summary 327
References 328
EPILOGUE 335
APPENDICES 337
A NUMERICAL QUESTIONS 337

References 34 1
B GLOSSARY OF TERMS 342
C IMPORTANT DEVELOPMENTS IN 'SUPERVOLTAGE'

RADIOTHERAPY FOR CONTEXTUAL FRAMING OF
CONFORMAL RADIOTHERAPY 354
References 356
INDEX 359

PREFACE
Conformal radiotherapy is tailoring the high-dose volume to the target volume in

a patient, simultaneously delivering a low dose to non-target tissues. By and large
the term is associated with extemal-beam radiotherapy although there is no reason
why this distinction should necessarily be drawn. Brachytherapy is in a sense the
ultimate in conformal therapy but has its own rich literature and is not covered in
this volume. What makes conformal radiotherapy so novel and exciting is that,
whilst the ambition to achieve it is as old as x-rays themselves, it is only in very
recent years that the prospect is becoming reality. This is because of the synergistic
confluence of progress in radiation acceleration technology, in computational
power for three-dimensional dose planning, in three-dimensional imaging for
tumour localization and therapy monitoring and in techniques for ensuring and
assessing the accuracy of positioning at the time of treatment. Developments
in automatic techniques for shaping the beam, tools for characterizing three-
dimensional dose distributions and methods of predicting biological response from
dose plans are all new and contribute to this synergism.
It might be argued that it is too soon to prepare a work reviewing this field. All
these aspects of conformal therapy are in a fluid state, continually being developed.
In this volume I have tried to extract the key elements which will probably stand
the test of time; but it is inevitable that conformal therapy will continue to develop
and this book is prepared in the hope of being useful during this development.
Some of what is presented is moderately controversial and where this is the case
I have tried to document the elements of doubt and debate.
Why select high-energy photons as the tool for conformal therapy? By the
very nature of photon-tissue interactions, they have an in-built disadvantage
in scattering and depositing doseproximaland distal to the target volume. In
an ideal world they would not be the radiation of choice for this challenging
problem. Protons are inherently ‘better behaved’ since, to first order, they travel in
moderately straight lines and they can be tailored to deposit none of their energy
distal to the target. Protons have been irradiating patients for close on 40 years
yet the worldwide patient totals are still only some lo4compared with millions of
photon treatments. Until now the technology has been simply too costly, yet now
this modality is receiving renewed interest and is reviewed here.
Conversely the unwanted photons distal to the target volume and exiting from
the patient can be used to advantage in the developing field of portal imaging.
ix
X Preface
Traditionally, film has been the medium of choice, but in the last ten years new
detectors have been developed with many advantages over film as detector, a
possibility obviously not available to proton irradiation whose quality control has
to be performed differently. Megavoltage imaging, both planar and tomographic,
is a growing activity, largely, though not exclusively, non-commercial-based at
present.
In the last ten years magnetic resonance imaging has been added to the arsenal
of imaging modalities which have been applied to planning and monitoring
radiation therapy. These include x-ray computed tomography, digital subtraction
angiography, single-photon emission computed tomography, positron emission
computed tomography and ultrasound. Organizing tomographic imaging into
truly three-dimensional treatment planning has become another major growth area
both within university and hospital centres and in industry.
Techniques for treatment plan optimization have been available for many
decades. In particular, optimization by rotation techniques were once very much in
vogue. By drawing analogies with image reconstruction methods, dose planning
by inverse computed tomography has attracted interest and become a very real
possibility with modem, fast digital computers. Using the multileaf collimator to
create beams with irregular geometric shapes, tailored to the projection area of the
target volume, it is now possible to optimize the arrangement of a finite number
of such fields, creating irregular and indeed concave high-dose volumes. The
multileaf collimator is over ten years old, but only now coming into use in more
than just a few centres. Photon dose planning based on point-dose kernels (the
distribution of dose around a photon interaction at a point) is also just becoming
possible.
Stereotactic radiotherapy and radiosurgery have been practised for four decades
but with the realization that linear accelerators can potentially do the work of
the special-purpose gamma knife machines, there is a resurgence of interest.
The underlying physics requires a very special confidence in the use of medical
images to guide the precise position of the small-area radiation beams. Along
with brachytherapy this is also in a sense the ultimate in conformation. It finds
use not only in treating tumours but also benign disease such as arteriovenous
malformations.
Many physicists and radiotherapists content themselves with dose distributions,
relying on clinical experience to translate these into the predicted biological
response of tumours and normal tissue. Work is underway to put the translation
from the surrogate, dose, to biological outcome onto a firmer footing. The field
is controversial, not least because the models are complicated and much of the
basic radiobiological data are poorly understood or disputed. Statements about
biological outcome made here should be treated with caution.
The process of treating a patient is a chain involving all these activities. Any
weak link in the chain reduces the efficacy of making the other steps more precise
(figure P. 1). Hence although most of us can only expect to contribute to one small
part of the process, a perspective on the wider context is needed. Some go SO far

Preface xi
positioning Dose evaluated

(tools, dose-volume histogram)
Figure P.1. Showing the 'chain' of processes leading to conformal

radiotherapy and illustrating that the physical basis of radiotherapy depends
on satisfactory pelformance at all stages.
as to say that an effective radiotherapy centre with its physics support must be able
to demonstrate that attention has been paid to all parts of the chain. This is not
easy in a world of limited resources, both human and material, but it should be a
goal.
Steve Webb
April 1992

ACKNOWLEDGEMENTS
The Institute of Cancer Research and the Royal Marsden Hospital collaborate
on a large research programme working towards improving the physical basis
of radiotherapy leading to achieving conformal radiotherapy. Clinical research is
conducted under Alan Horwich in the Department of Radiotherapy and, alongside,
the Joint Department of Physics under Bob Ott is developing new techniques
to improve the physical basis of conformal radiotherapy. In tum, this research
fuelled interest in a postgraduate teaching course on the physics of conformal
radiotherapy, which led to the production of this book. I am enormously grateful
to all those whose enthusiasm for this subject has supported this research and
teaching.
I am particularly grateful to my colleagues who are experts in specific aspects
of radiotherapy physics who have shared their knowledge with me. These include
Bill Swindell (Head of the Radiotherapy Physics Team), Alan Nahum, Philip
Mayles, Philip Evans, Mike Rosenbloom, Roy Bentley, Glyn Shentall, Margaret
Bidmead, Jim Wamngton, Sarah Heisig and Penny Latimer. Several PhD students
including Glenn Flux, Michael Lee and Rong Xiao, who attended the course, have
also provided helpful comments.
I am most grateful to Sue Sugden and Pru Rumens in the Institute of Cancer
Research Library for helping me to obtain copies of references and to Ray Stuckey
and his photographic staff for help with half-tone illustrations.
The research work in conformal radiotherapy is concentrated in so few
centres worldwide that in writing these reviews, some ‘household names’ in the
radiotherapy physics world keep appearing. It would be invidious to single out
some subset for listing here but the reader will have no difficulty identifying them
from the fruits of their research and my greatest debt in preparing this work is
to them. I hope this compendium will make their work even better known and
that they will approve, even though inevitably this can be only a prtcis of their
achievements.
I should like to thank all the publishers who have allowed figures to be
reproduced. The authors are acknowledged in the figure captions. All publishers
were contacted with request for permission to reproduce copyright material.
The reference lists are up to date as of May 1992.
...
Xlll

xiv Acknowledgements
The work of the Institute of Cancer Research and Royal Marsden Hospital is
supported by the Cancer Research Campaign.
I thank Sean Pidgeon (CommissioningEditor) at Institute of Physics Publishing
for his enthusiasm for this project.
I am also grateful to Mark Telford and Jenny Pickles of Institute of Physics
Publishing for their excellent work in the publication of this book.
This book is dedicated to Linda and the fight against chronic disease.

CHAPTER 1
THREE-DIMENSIONAL
RADIATION-THERAPY TREATMENT
PLANNING
1.1. CONFORMAL RADIOTHERAPY TREATMENT PLANNING
1.1.1. What is conformal radiotherapy?

The aim of radiation therapy is to tailor a tumourcidal dose envelope to a target
volume and to deliver as low a radiation dose as possible to all other normal tissues.
This is what is meant by conformation or conformal radiotherapy-the high-dose
volume ‘conforms’ to the target (figure 1.1). By target volume is implied the full
extent of the tumour including any marginal spread of disease (sometimes called
the biological target volume (BTV)) and a ‘safety’ margin (extending the BTV to
the so-called mobile target volume (MTV)) (Une et a1 1991). A simultaneous
goal is to minimize the dose to organs at risk. Against the background of
these aims, what is perhaps surprising is that so little practical radiotherapy
completely achieves this goal. The reasons are not hard to establish. Difficulties
in determining the three-dimensional target volume, in designing appropriately
shaped and oriented radiation ports, in calculating the dose distribution from
complex superpositions of radiation fields in inhomogeneous tissue and in
establishing that the patient is correctly positioned at the time of treatment
delivery, all conspire to defeat the aims. It is now recognized that only by
improving the physical basis of radiotherapy at all of these stages can the problem
of accurately establishing the physical basis of radiotherapy be said to be properly
tackled. As may be expected there are several groups of workers attacking separate
aspects of the problem but few taking on the full burden of solving the complete
chain of tasks (Schlegel 1987, Suit and Verhey 1988). In the meantime practical
radiotherapy is a patchy activity varying at worst from techniques little changed
for decades, through to the most ambitious protocols, with a wide spectrum
inbetween. Optimizing radiotherapy with the goal of conformation is highly
desirable to remedy this situation. This is not meant to imply that when the
physical basis of radiotherapy has been optimized all the problems magically
1
2 Three-dimensional radiation-therapy treatment planning
Figure 1.1. Illustrates the difference between ‘conventional’ and ‘confor-

mal’ radiotherapy. In the upper figure a rectangular beam is shown irradiat-
ing a target volume (shaded)which changes its cross section with longitudinal
position. Radiotherapy might comprise a number of such beams, coplanar,
with at best some blocking. Inevitably this irradiates some normal tissue in
an unwanted manner. In the lowerfigure, one of many, possibly non-coplanar,
beams is shown with the shape of the beam tailored to the projection of the
target in that direction. The result of using a number of such beams should be
a dose distribution (an isodose surface is represented dotted) which conforms
to the sutface of the target volume.
vanish. There will still be difficulties associated with incomplete radiobiological

knowledge-i.e. how does tissue damage correlate with radiation dose-and of
course this whole discussion can only apply to treating well-differentiated local
primary disease; metastases are another matter (Horwich 1990).
It has been estimated (Suit and Westgate 1986) that improved local control is
so important that thousands of lives per year could be saved by improving the
physical basis of radiotherapy and this has become the driving force behind the
efforts to achieve conformal therapy.
Smith ( 1 990) voices concems about how the new technology for conformal
radiotherapy should be evaluated. In his view many authors have not
reported quantitative evaluations of the real clinical importance of their physics
developments. In particular whilst claims that conformal therapy reduced
the dose to normal tissues, it was felt necessary to prove that the tissue
otherwise irradiated in non-conformal therapy was clinically damaged. Conformal
techniques should be compared with the best available non-conformal techniques.
Techniques improving dose calculation should be quantified in terms of tumour
control probability and normal tissue complication probability with the dose

Conformal radiotherapy treatment planning 3
normalization changed to increase the latter to tolerance and consequently improve

the tumour control. Attention should be given to the possible problem of ‘good’
dose distributions being spoilt by unwanted patient movement.
Conformal radiotherapy is technologically complex and would appear to be
justified by evidence that improved dose distributions would lead to improved
local control. Lichter (1991) argues that these technological improvements should
now be given the opportunity to show whether real clinical advantages can be
produced compared with conventional technology. He raises several, almost
moral, dilemmas. Firstly, researchers should take time to define the clinical
experiments, the end points and the potential outcomes. Are we interested in
more cures or better local control with reduced morbidity, or the same cure
with fewer acute side effects? Secondly, Lichter questions the timing of such
trials. Should these perhaps wait until the technology of multileaf collimators
(chapter 5 ) , optimization (chapter 2), megavoltage imaging (chapter 6) and 3D
treatment planning are all well established? Thirdly, would patients tolerate being
randomized into low-technology versus high-technology trials?
Since three-dimensional treatment planning (3DTP) is central to conformal
therapy we begin with discussing planning tools.
1.1.2. The planning tools required for conformal radiotherapy

Fraass (1990) and Fraass et al (1990) have broken down the planning problem
into its component physical considerations. The improvement offered by the most
recent treatment planning computational tools arises from the use of fully three-
dimensional data to define anatomy, design beams, calculate doses and evaluate
plans. These tools are now becoming available in some centres as computational
hardware becomes fast enough and as the problems of utilizing three-dimensional
medical images are solved. Some three-dimensionalplanning systems have been
developed ‘in house’ by university hospitals and commercial three-dimensional
planning systems are also becoming available.
1.1.2.1. Conventional 2D treatment planning. Let us recall what we might call

the ‘conventional’ approach to treatment planning (and in doing so make some
broad generalizations). Rosenman (1 99 1) provides a useful thumbnail historical
account of the development of radiotherapy planning and the subject is given
tutorial coverage by Bleehan et a1 ( I 983). Before the advent of commercially
available computerized tomography (CT) around 1972, the target volume was
estimated from planar radiographs. With the patient lying in the treatment
position, a few radiation fields or ports were determined at the simulator with
the gantry at maybe two, three or four orientations spaced on the circumference of
a single circle. The ports could be of different sizes at each orientation but were
generally rectangular in shape since this is all the field collimators could achieve
(figure 1.2). The use of a single circle was regulated by the treatment machine,
which could only deliver coplanar fields. Non-coplanar fields were possible by

Figure 1.2. An illustration of some of the features of ‘conventional’ (i.e.

non-conformal) radiotherapy. The patient is being treated with a pair of
parallel-opposed lateral x-ray fields and an anterior-posterior field. The
three fields are coplanar, i.e. their central axes all lie in a plane with the
fields being set up by rotating a gantry about a stationary patient lying on
the patient support system. Each field is collimated to a rectangle by two
pairs of orthogonal lead jaws. Ci ( i = 1, 2 , 3 ) represents the collimation.
The high-dose volume, where the three jields overlap in the patient, is
approximately a parallelpiped (not exactly because of beam divergence).
Unless the tumour happened to be this shape, which is unlikely, much normal
tissue is unnecessarily irradiated. Even if it were this shape, the dose from
just a few beams would have inhomogeneities. The fields might be partially
blocked to change this shape. Pi (i = 1 , 2, 3) are the three exit portals.
Conventionally a film will record the exit radiation and be compared with a
simulator film. The portals are not shown equidistant from the isocentre to
avoid cluttering the diagram. In practice the film would be placed closer to
the patient than shown. The treatment couch is flat.
twisting the couch but treatment planning was difficult if dose calculation was
not confined to a single plane. The assumption was made that the patient was
cylindrically uniform over the axial extent of the radiation fields. An outline of
the extemal contour was made in the midplane of the fields using a lead wire or
a plaster-of-paris cast. A treatment plan was calculated-in early days by hand
but later by computer-using this outline, and appropriate orientations and beam
weights for the ports were found. The dose plan was generally represented by a
series of isodose lines in this single transaxial plane. A decision on the suitability
of the plan was made by eye based on rules of thumb and experience.
Of course the above description is arguably too simple but serves as the basis
of explaining the improvements which are now being introduced with three-
dimensional treatment planning. Decades of experience honed a variety of
techniques tailored to specific tumour sites and ingenious attempts were made to
improve matters. Before CT, simple blurred ‘classical’ tomograms were used to
visualize transaxial anatomy; couch and head twists were designed to avoid the
restriction to coplanarity; estimates of air spaces were made and simple tissue-

inhomogeneity corrections were attempted. Many hours at the planning computer

were devoted to trying differentbeam positions to avoid sensitive structures and to
improve the dose distributions. Blocks were designed at the simulator. However,
all up, the available technology provided the limit which no amount of human
ingenuity could overcome.
Let us critically list the limitations of the classical approach:
1. Planar x-rays often did not show clearly the tumour extent.
2. The tumour extent was determined anatomically-functional imaging with
good spatial resolution was often unavailable and, where available, was often not
correlated with anatomy.
3. The correct 3D shape of the tumour was not determined.
4. The 3D location of sensitive structures was not determined.
5 . The assumption that the patient was uniform axially over the length of the
treatment field was wrong.
6. Planning was made in a single axial slice, usually a mid-field slicethere
was no distinction between the behaviour in this slice and that in adjacent slices.
7. The use of a few beams limited the possibilities for tailoring the dose to
the target region. The high-dose volume was far larger than the target volume
and contained normal and potentially sensitive structures-so-called ‘organs at
risk’. Tumours with irregular shapes were ‘boxed in’ to cubic or rectangular-
parallelpiped high-dose volumes.
8. The beams were axially coplanar.
9. The body contour was difficult to determine-it may even have been altered
by the method of measurement (e.g. pressed by lead wire).
10. The internal tissue inhomogeneities were hard to determine accurately.
11. The dose plans were computed by an algorithm which was too simple.
12. The dose planning took no account of the 3D nature of the body surface and
changes in density in 3D.
13. The doses were displayed as isodose lines in a plane. There was no
quantitative method of determining how much of the tumour received dose within
specified ranges. Even if this were worked out for the single plane in which
the doses were computed there was no way of finding out what happened in the
adjacent tissue.
14. The beam ports were rectangular, not tailored to the projection of the
treatment volume.
. . .and so one could go on. Perhaps the most serious deficiency was that one
simply had no way of knowing what dose had really been delivered.
Notice in this catalogue of difficulties that there were several ‘missing
technologies’ which all conspired to produce these limitations. Medical
imaging was planar and largely anatomical. Treatment machines had very
limited geometrical movements (rotational movements and rectangular fields).
Computers were too slow for accurate dose planning (see, for example,
Cunningham 1988) and graphic display of dose and tissues was limited.

1.1.2.2. The impact of x-ray CT. In 1972 one of these limitations was
dramatically removed with the advent of x-ray computed tomography (CT).
In a very short time, accurate 3D tomographic information on both diseased
and normal tissues became available and was quickly harnessed to treatment-
planning computers (Parker and Hobday 1981, Goitein 1982, Lichter et a1 1983).
However, no such widely available dramatic improvement followed in the design
of treatment machines, which were still only able to deliver rectangular (maybe
blocked) fields from a limited number of angular directions. There were still no
3D computer planning systems. So by no means all the limitations listed above
disappeared. It was still the norm to make use of a mid-field axial slice and
imagine the patient to be uniform axially for the purposes of treatment planning.
However, it was possible to superpose the central slice plan on the true anatomical
adjacent slices and make some qualitative observations. There was generally
no attempt to compute the dose-volume histogram (i.e. the histogram showing
the fractional volume of tissue raised to specific dose levels-see section 1.1.9),
partly because, of course, the non-central dose distributions were incorrect and
partly because such calculations were still prohibitively time consuming. In short,
3D dose calculations were generally not performed despite the availability of 3D
anatomical information. 3D display and plan evaluation had to be sacrificed as a
result. The limitations of the treatment machines prohibited imaginative choices of
port shapes and the use of a large number of beam orientations. It is for this reason
that, although in terms of visualization, CT had a dramatic impact on radiation
treatment planning, the methods used were still essentially two dimensional. Most
of these requirements for 2D imaging were met by so-called ‘tum-key ’ systems,
marketed commercially, widely tested, but somewhat limited in scope.
Lichter (1986) has pointed out that the clinical results obtained using CT-based
treatment planning have never been evaluated in comparison with conventional
planning methods for the very good reason that CT enabled clinicians to define
anatomy so much better that it would have been unethical not to use the data in
one arm of a trial (data on the frequency of changing plans with CT knowledge is
available but this is not the same as assessing outcome).
1 .I .2.3. Three-dimensional multimodality medical imaging and 3 0 treatment

planning. Several attempts were made soon after CT data became available to
use it properly to compute 3D dose distributions. However it is only now that
computers are sufficiently fast to allow this in sensible times. With the passage
of time since the early seventies, further 3D imaging modalities have become
available, such as emission tomography (SPECT and PET) and magnetic resonance
imaging (MRI). For an overview of the physics and engineering of 3D medical
imaging the reader could choose from (among other books) Moores et a1 (1981),
Hamilton (1982), Wells (1 982), Gifford (1984), Guzzardi (1987), Aird (1988) and
Webb (1988). Aspects of medical imaging relevant to planning and evaluating
radiotherapy are discussed in chapter 8 along with some brief details of how these
technologies work. There are special requirements for CT scanning producing

images for radiotherapy treatment planning (Lichter et all983, McCullough 1983,

Goitein 1983, Sontag and Purdy 1991) including: a flat couch, patient in the
treatment position with quiet respiration, contiguous thin slices (3-5 mm in the
head; 5-10 mm in the body), large patient aperture, scout film indexed to scans
etc. MRI allows imaging in arbitrary planes, can differentiate certain tissues
better than CT and is useful for monitoring response to treatment, but may suffer
image geometrical distortion and does not give electron densities directly (Sontag
and Purdy 1991). Moerland et a1 (1991) have suggested distortion could be as
much as 5 mm. Ultrasound does not find great use except in determining chest
wall thickness and external contours. Emission computed tomography is also
beginning to play a part in treatment planning and monitoring. Techniques to
merge and combine 3D image data sets are being developed (see section 1.2);
non-axial slices may be viewed. Using the beam’s-eye-view of 3D structures
created from 3D image datsets, multiple non-coplanar geometrically shaped fields
can be defined at arbitrary orientations (figure 1.3) (McShan et a1 1990). 3D
dose calculations can be performed including 3D inhomogeneity corrections and
the results may be shown superposed on 3D anatomy either as sequential slices
or as shaded-surface distributions. These facilities enable development of tools
for plan analysis and comparison whereby it can be precisely established what
fraction of the treatment volume receives dose in prescribed ranges. The scene
is set for removing all of the physical limitations listed in section 1.1.2.1 and
to permit true 3D planning (Sontag and h r d y 1991, Laughlin et al 1991). Two
major obstacles remain. Firstly, the computational hardware is expensive and the
software environments to achieve these goals are only developed at a small number
of centres. Some are available commercially, but at a high price. For the most
advanced dose-calculation techniques, computer times remain high. Secondly, the
treatment-delivery machine hardware has simply not kept up and is still geared to
the ‘conventional’ era or at best the ‘2D planning with CT’ era. Some imaginative
workers are developing dynamic therapy (see chapter 7) or multileaf collimators
(MLC) and there is a need to automate the computation of the settings of the leaves
(see chapter 5 ) . In the early 1990’sthese are not yet the normal modes of treatment.
At least in a few centres, planning treatments with full use of three-dimensional
information has been achieved and attention can tum to questions of plan
evaluation. Once again the tools may be available to quantify dose-volume
histograms, but there is still considerable debate over how to tum this physical
data into a prediction of biological response (Tait and Nahum 1990) (see section
1.1.11). This requires knowledge of tumour control probability and normal tissue
complication probability, and whilst some data are available (e.g. Lyman 1985),
and methods of using it (Kutcher and Burman 1989), the data are incomplete and
open to questioning (McShan 1990b). Against this background it is not surprising
that most workers still interpret the treatment plan in terms of dose assessment
rather than biological effect. Coupled with this is the development of methods
to ensure that the patient is correctly positioned at the time of treatment. Digital
portal imaging (see chapter 6 ) and megavoltage computed tomography can gather

Figure 1.3. An illustration of some of the features of ‘conformal’

radiotherapy. The patient is being treated by a number of beams which have
been shaped to the projected area of the target by a multileaf collimator ( M ) .
This enables the fields to departfrom rectangular and take an irregular shape.
The central axis of the one beam shown does not lie in the transaxial plane of
the patient because the longitudinal axis of the couch (C) is not orthogonal
to the plane of rotation of the gantry ( G )(couch twist). One might imagine a
number of beams whose central axes do not lie in a single plane, a situation
described as ‘non-coplanar’. The image of the portal radiation is captured
by an electronic portal imager ( E ) diametrically opposite the x-ray source.
the data to assess how well this is done with present positioning techniques and
allow repositioning, should errors be thought excessive (Wong and Purdy 1990,
Evans eta1 1990, Wong 1991).
If the 2D distribution of portal intensity is measured by a digital scanner, the
intensity can be backprojected into the megavoltage CT imaging data (Ishigaki et
a1 1990). This gives a first-order map of primary dose. If the dose kernel (see
chapter 2) were to be folded into this, the result would be a map of delivered dose
including scatter. Such ‘transit dosimetry’ is still in its infancy and not a technique
in the repertoire of most radiotherapy departments (see also chapter 6, section 12).
It would not be particularly instructive in a teaching text to describe in detail
specific treatment-planning systems since these are continually developing and
not all the features required of a 3D treatment planning system are available in all
systems. Also this level of detail is only really needed by the user of a particular
system who will want to consult the papers of the groups concemed. The main
3DTP systems currently in use are at:
0 the Department of Radiation Oncology, University of Michigan at Ann Arbor
(McShan and Fraass 1987, Fraass and McShan 1987, Fraass 1986, Fraass
et a1 1987, McShan 1986, 1990a,b, Kessler et a1 1991) (*SCANDIPLAN;
Scanditronix);
0 the Joint Centre for Radiation Therapy, Massachusetts General Hospital, Boston

(Goitein et a1 1983b, Kijewski 1987);

0 the German Cancer Centre at Heidelberg (Schlegel et al 1987, Boesecke et al
1987) (*part system/ STP, Fischer);
0 the University of North Carolina at Chapel Hill (Sherouse et a1 1990);
0 the University of Tampere, Finland (Vitaenen) (*CADPlan; Varian);
0 the Karolinska Hospital, Stockholm, Sweden (Brahme 1992);
0 the Memorial Sloan Kettering Cancer Centre, New York (Laughlin et a1 1991);
0 the University of Pennsylvania School of Medicine and Fox Chase Cancer
Centre, Philadelphia (Sontag et a1 1987, Sontag and Purdy 1991);

0 the Mallinckrodt Institute of Radiation, Washington University School of
Medicine, St Louis (Purdy et a1 1987, Sontag and Purdy 1991).

The systems asterisked have been marketed commercially (names shown in
brackets). Other commercial systems include the AB Helax system.
Instead of describing these 3D planning systems in detail, the principal
requirements for 3DTP and how these might be met are discussed in this chapter.
Sontag and Purdy (1991) emphasize the need for good documentation for new
3D treatment-planning systems. Jacky and White (1990) have pointed out the
requirement for careful validation of the accuracy of 3D planning software. By
comparing the predictions of computer calculations and the mathematics which is
supposedly coded, the programs are validated independent of the weaknesses of
any particular beam model. Verification by not comparing with measurements can
show up programming errors which might otherwise masquerade as inaccuracies
of the model. Smith (1990) summarizes the two major initiative in 3D radiation
therapy planning covered by the Nordic CART (Computer aided radiotherapy)
initiative, started in 1985 and by the American Centres supported by NCI contract,
started in 1982 and reported in a special issue of the Intemational Joumal of
Radiation Oncology, Biology and Physics in May 1991 (see also section 1.3).
Brahme (1 992) has re-emphasized that fully 3D treatment planning requires
all of the following to be performed three-dimensionally: diagnosis, biological
response monitoring, dose calculation (including optimization and possibly the
use of 3D dose kemels), treatment simulation, display, dose delivery, fixation and
dose verification.
Achieving 3D treatment planning does not necessarily in itself imply that
conformal therapy will be the result. A poor set of choices for the beams would
certainly not achieve this objective. However, 3D treatment planning is such
an important component of the chain of events leading hopefully to conformal
therapy, that it is the first part of that chain that is discussed in some detail.
I .I .3. The computer system

Among the earliest 3D planning systems making use of extensive computer
graphics were those of Reinstein et a1 (1978) and McShan et al (1979). As
computer facilities have become faster and cheaper, more sophisticated features
have been developed. A 3DTP system requires a computer with a large storage

capacity (several gigabytes) for 3D images and dose distributions together with
some hundred megabytes of memory. Graphics display systems supported by
device-independent GKS graphics software need to be attached. These should
preferably support multiple-window displays, since many people find it easier to
build up a 3D impression from simultaneously viewing multiple 2D slices than
from true (shaded-surface) 3D displays of anatomy and dose. DEC VAX systems
were chosen by the Ann Arbor team for example to meet these requirements. The
system should be menu driven with interaction by mouse and keyboard. 3D dose
calculations take a long time and for these batch computing facilities are needed
(Tepper and Shank 1991).
1.I .4. Localization and coordinate systems

Functional imaging modalities such as SPECT and PET, and anatomical imaging
by x-ray CT and MRI can all provide data for input to the treatment-planning
process. There is no real concensus yet on the impact and usefulness of MRI data
in treatment planning, but this will surely come in the next few years. Margulis
(1992) has stressed that medical images, including MRI, assist accurate staging
of disease as well as the determination of the target volume. Unfortunately the
data obtained from these imaging modalities are not usually in the same format.
Pixel/voxel sizes, the number of pixels per frame (tomographic slice), conventions
for labelling and sequencing frames, are all potentially different. Also the patient
may not have been in the same position when each of the studies was performed.
Consequently the first requirement of a treatment-planning system is that it should
be able to accept a variety of data input types and reformat data into a common
arrangement of voxels. Secondly, it should provide for converting these into a
single common reference system. The first requirement is not difficult to meet
provided the manufacturer of the imaging equipment will divulge the data format.
The second is far more difficult. Rotations, translations and changes of scale
are relatively easy to accomodate by matrix algebra, but these only apply if the
relative movements between datasets can be considered rigid transformations. If
one set is warped relative to another the problem is more complicated and many
techniques have been tried to overcome this (e.g. Kessler 1989). In view of
the fundamental importance of image registration in three-dimensional radiation
treatment planning, this topic is treated at length in section 1.2. The end result of
these transformations is that the image capability of the treatment-planning system
should become independent of the origin of the imaging studies. The planning
system should use a coordinate system tied to the treatment room and the images
should be locked into this system. Fraass (1986), Fraass et a1 (1990) and McShan
and Fraass (1990) have called this the ‘dataset concept’ and implemented it in
the University of Michigan planning system ‘UMPLAN’. The use of coordinate
transformations also allows positioning differences between the simulator and any
single imaging modality (such as the CT scanner) to be observed and corrected. If
imaging is available on the treatment machine then positioning errors at the time

of treatment could be similarly corrected (see chapter 6). The imaging coordinate
systems may be defined relative to some marks on the patient’s skin or attached
markers at the time of scanning or with reference to bony landmarks or external
patient restraint systems (see section 1.2).
1.1.5. Structuresfrom images

The staple input is x-ray CT, generally a large number of transaxial slices together
with a scout view (see also chapter 8). The scout view is a digital planar x-
radiograph taken on the CT scanner without gantry rotation and showing the
longitudinal positions of the transverse sections (which are sometimes called slices
or cuts) (figure 1.4). It is not the same as a digital planar radiograph taken with
the beam diverging in all directions; in the scout view the beam diverges only in
the transaxial direction. The treatment planner should be able to request non-axial
slices including oblique cuts. Structures can be outlined by autotracking, i.e. the
computer will automatically threshold the CT numbers to create a contour. Where
this is inappropriate the contours of structures may be drawn by hand, usually
using a mouse or trackball. If a large-screen digitizer is available, this process is
more accurate. In particular, parallax errors can be reduced (McShan et al 1987).
3D treatment planning requires large numbers of contiguous CT slices and there is
clearly a tradeoff between wanting thin slices and wanting high patient throughput
on the CT scanner (Sontag and Purdy 1991). From the ensemble of contours, 3D
solid surface views can be created. Some (e.g. Gildersleve 1991, Wong and Purdy
1990, Tepper and Shank 1991) regard this aspect of treatment planning as the most
time consuming, generating a bottleneck in the series of processes from imaging
to treatment setup. There is thus growing interest in ‘image segmentation’ and
also in ‘artificial intelligence’, whereby a computer might be able to recognize
specific anatomy from its shape, density or location in relation to other structures
(Sontag and Purdy 1991). This has presently not been achieved and would be
a difficult computer problem to solve. The basic need is to define (i) the ‘target
volume’ being the tumour volume, margins of spread, safety margins, in fact all the
tissue which it is required to raise to high dose, and (ii) ‘organs at risk’, being the
sensitive structure which must be kept at low dose. If these are next to each other
high dose gradients are the aim. There is plenty of evidence that the most important
strategy is to achieve superior dose distributions in terms of increasing tumour
control probability and also reducing treatment related morbidity. Obviously the
worst possible error is to improperly define the target volume; this can never be
corrected by later adjustments. For this reason conservative estimates of the target
volume are often made.
Wambersie et a1 ( 1988) stress the over-riding importance in treatment planning
of correctly determining the target volume. This is a clinical decision and no
amount of quality control of other parts of the chain of activities involved in
delivering radiotherapy can correct for wrong specifications at this stage.

c
Figure 1.4. Radiotherapy planning ( R T P ) makes use of the three-dimensional

imaging modalities, x-ray computed tomography (cT), magnetic resonance
imaging ( M R I ) , single-photon emission computed tomography (SPECT) and
positron emission tomography (PET). X-ray CT was historically first used in
RTP and is still the most common. Tomograms (slices)are made over a range
R spanning the expected location of the tumour T. P is the plane of the scout
view. An outline of the target T is made on each slice C. Other normal tissue
and organs at risk may also be contoured, together with the patient external
outline. These dataform the geometrical input to treatmentplanning. Oblique
slices may be made by reformatting transaxial data. MRI is capable of directly
generating non-transaxial images. 3 0 imaging may also be used to monitor
the outcome of therapy.
1.1.6. Beams
A truly 3 D W facility will allow beams to be specified from any orientation, be able
to block such fields and then show the resulting dosimetry superposed on any cut
through the 3D images. This requires that if multiple windows are in operation and
the beam positions are changed in one window (say, a transaxial cut) the beams
are immediately updated in the other windows displaying different cuts.
Sailer et a1 (1990) have provided a very neat solution to the problem of labelling
beams which can lie anywhere in 3D. Their paper is entitled ‘3D beams need 3D
names’. Radiation beams have traditionally taken their names, such as anterior,
posterior, lateral etc, from radiology. These names are unambiguous. However
extensions such as ‘oblique’ or ‘left anterior oblique’ fail to convey ‘how much
left?’ and ‘how much anterior?’ The solution lies in using the three primary axes in
the patient. These are left(Ljright(R), anterior(Ajposterior(P), and inferior(1)-
superior(S). Generally oriented beams can then be labelled: A 101A&A3 where
A 1 is the nearest primary axis to the axis of the beam, 81 the angle to the second-
closest primary axis A2 and & the angle to the third-closest primary axis A3.
For example the beam shown in figure 1.5 is ‘LAOA201’.By definition the larger
angle must appear first in the specification and this is a useful check. Beams

a5400
b-200
inlniar
Figure 1.5. A suggestion for how the directions of beams might be labelled.
A ‘complex beam’ (i.e. one not along a principal axis) is located in one octant
of three-dimensional space. This beam, which is closest to the left principal
axis, is named ‘L40A201’ (from Sailer et a1 ( I 990))(reprinted with permission
fiom Pergamon Press Ltd, Ogord, U K ) .
lying in a primary plane-one of transaxial, sagittal or coronal-have a simpler

three-element specification. For example a beam labelled ‘A20L‘ lies in the
transaxial plane, whilst ‘A30S’ lies in the sagittal plane. An example of a ‘simple
beam’ lying along an axis would be ‘L‘.This method should give unambiguous
communication of field directions, relative to patient axes. The transformation
between patient axes and equipment (gantry/couch/collimator) axes would be
needed to calculate the beam orientations in these coordinates. The terminology
was extended by Goitein (1990).
I .I .7. Blockedfields and the beam’s-eye-view

The beam’s-eye-view (BEV) of the target geometry is particularly useful for
making decisions conceming the shape of blocks or irregular multileaf-collimator-
defined fields to shield sensitive structures (figure 1.6). The concept was
introduced by Goitein er a1 (1983a). A popular form of BEV is to display a
perspective view, from the beam’s source position, of stacked contours (figure
1.7). This is especially effective when depth cueing (closer regions made brighter
and vice versa) is used. Shaded surface BEV display is also possible (Tepper and
Shank 1991). Sontag and Purdy (1991) emphasize the improved usefulness of the
BEV if it can show the isothickness lines of any wedge in the field. The shape of
blocks should be able to be drawn directly on the BEV display using a joystick or
trackball. This can sidestep the practice of defining blocks at the simulator which
might only be changed when verified at the BEV stage. Block design becomes
part of the treatment planning rather than the simulation process. The use of
the beam’s-eye-view can also effectively determine the optimal direction of beam
entry (Wong and Purdy 1990). Where the flexibility exists it is obviously good to
choose views of the target volume which exclude as much nearby sensitive tissue
as possible (Reinstein et a1 1983). Indeed, Goitein (1982) argues that determining

Figure 1.6. Schematic showing an x-ray beam collimated (non-conformally)

to a rectangular cross section intersecting a patient. T represents the target
volume; 0 represents an adjacent organ at risk; C is a plane orthogonal to the
beam axis at which the ‘beam’s-eye-view’of the patient may be constructed.
This (as the name suggests) is the view of the patient from the source-it is
shown schematically connected to C with a dotted line. Note in this example
the view of the target volume overlaps the view of the organ at risk. It might be
possible to remove the overlap by orienting the beam differently. A conformal
field would shape the radiation to the outline of the target as seen at C by a
multileaf collimator (see chapter 5). P represents the plane in which a portal
image can be formed (see chapter 6 ) .
the best geometrical field shape (to exclude as far as possible organs at risk)
is a matter of higher priority than paying undue attention to the algorithm for
calculating the dose to the target volume. This philosophy explains why many
commercial treatment planning computers put great emphasis on geometrical
precision and still use relatively uncomplicated photon algorithms. The use of
non-coplanar beam portals is clearly fundamental (Laughlin et a1 1991). It tums
out that the optimum set of geometrically shaped beams defined by multi-leaf
collimators (see chapter 5 ) can be determined entirely computationally so as to
achieve this aim (Webb 1991; see also chapter 2). The process of planning by use
of beam’s-eye-view is sometimes called ‘virtual simulation’ (Sherouse et a1 1990,
Sherouse and Chaney 1991).
The 3D radiation therapy treatment planning system at the University of North
Carolina is known as the ‘virtual simulator’ (Rosenman 1991). The essential
feature of this 3D treatment-planning system is that all planning is done on the
computer, based on tomographic (CT) data. A screen displays wire-frame 3D
images of the patient and small versions of selected CT slices are also on screen.
As the planner manipulates the beams looking at the 3D image, the positions
of the beams in the 2D images is continually updated. When the final choice
of beam positions has been made and the dose calculation done, the beam’s-
eye-view and portal radiograph for each field (sometimes called the ‘digitally
reconstructed radiograph (DRR)’) are computed (figure 1.8). Only then does
the patient go to the physical simulator for real x-ray images to be made for
quality assurance purposes. Some other treatment planning computers (e.g. IGE
TARGET and THERATRONICS CT-SIM) also have this facility. The DRR
becomes the standard against which portal images taken on the treatment machine

Figure 1.7. The beam’s-eye-view of the target T may be derived from CT

data. Each slice allows one contour to be defined. In the beam’s-eye-view V of
the target the contour is seen edge-on and can be represented by a thin looped
line. Each slice gives rise to one thin loop so in the example shown 17 slices
span the target. The brainstem B can also be seen in the beam’s-eye-view. It is
clear in this example that a rectangular field enclosing the target would also
enclose part of the brainstem. The beam’s-eye-view thus allows the treatment
planner to define the position of blocks. Alternatively thefield could be shaped
by a multileaf collimator. Because of beam divergence each contour in reality
appears in perspective as a thin closed curve. Only the central axis of the
beam sees the contour as a line. For simplicity single lines have been shown
here instead.
can be compared (Tepper and Shank 1991). However, it is resolution limited

because of the fairly large slice thickness of the CT data from which it is generated
(Sontag and Purdy 1991).
McShan et a1 (1 990) discuss the great importance attached to the beam’s-eye-
view facility in the UMPLAN 3D treatment-planning system. By establishing
the matrix transformation between the treatment reference system and the beam
coordinate system, structures can be mapped from the beam’s-eye-view projection
space to the tomographic image space and vice versa. Hence blocks can
be designed at the planning stage as in the ‘virtual simulator’. A digitally
reconstructed radiograph can be matched to the portal image. Indeed the DRR
can sometimes provide the only portal image in circumstances where the practical
portal imaging system would collide with the couch.
The beam’s-eye-view can also provide a measurement of the volume of normal
tissue irradiated by a portal, even when it is shaped to the projected area of the
target volume; this will occur if some or all of the organ-at-risk is also in the line
of sight of the irradiation. Chen et a1 (1992) have created plots of such a volume

Figure 1.8. A truly j D treatment planning system can generate a digital

portal radiograph (DPR) D by ray tracing from the x-ray source S through
the CT data C. This ray tracing takes account of the ray divergence
and differential photon attenuation. A series of CT slices are shown in
the pelvic region of a patient. Because the in-plane pixel size will be
smaller than the slice width, some interpolation is needed to obtain square
pixels in the DPR,sometimes also called a digital reconstructed radiograph
(DRR).The simplest DRR attenuates the primary radiation along each ray
path. An experimentally determined portal radiograph at the simulator has
contributions from scattered radiation also. Either the simulator radiograph
or the DRR can be compared with a portal image at the time of treatment on the
linear accelerator. If this is also digital, the comparison may be quantitative.
(or fractional volume of the total organ-at-risk) as a function of beam orientation

for a very large number of potential orientations including non-coplanar portals.
They make the very plausible argument that these plots can help select suitable
views. They do not claim this is an optimization technique since no doses have
been calculated at this stage of the planning process. However this method of
‘volumetrics’ can be a good way to select an initial set of orientations from the
large number of possibilities and then beamweightd wedges etc can be optimized
subsequently (see chapter 2). Another proposal from Chen et a1 (1992) is to
display ‘bitmaps’ showing what fraction of the volume of an organ-at-risk receives
primary dose from m fields when there are n > m fields in all.
The BEV, useful though it is, has one major limitation. It shows the
perspective of just one beam at a time. Sontag and h r d y (1991) have discussed
the possibilities for ‘physician’s-eye-view’(PEV) whereby wire frames for the
anatomy and all the beams are displayed as if the observer’s eye is in the treatment
room. PEV would help evaluate field abutments and gaps.
1.1.8. Dose calculation

The ideal dose calculation method should:
1. include all 3D geometry information;
2. make use of an accurate 3D map of electron density;

3. use a 3D description of the beam:

4. include 3D scatter and interface effects within the patient:
5. calculate the dose at all points in the volume;
6. be fast.
There is still no agreement on the ‘best’ photon algorithm (Cunningham
1982, Laughlin et a1 1991). Computationally requirements 5 and 6 clash.
Generally a crude matrix of dose points in selected planes are calculated fast
allowing scope for changes and optimization. Once it is believed the beams
are optimized, full fine-matrix calculations can be made. Perhaps surprisingly,
for photon dose computation, the majority of commercial planning systems still
use methods that were developed many years ago (Cunningham 1982, Sontag
and Purdy 1991). The subject of dose optimization is considered more fully
in chapter 2. It is now just becoming accepted that dose calculation based on
the integration over dose-kemels is the most accurate way to properly compute
three-dimensional distibutions (see chapter 2). However, elaborate though such
methods be, it is presently not possible to perform such calculations as fast as
needed for routine treatment planning. Hence for the time being they are generally
reserved for benchmark computations, leaving day-to-day calculations still based
on suboptimal dose-calculation models (Wong and Purdy 1990).
It is important to separate the problems of determining the dose level from the
dose distribution. Obtaining the most uniform distribution of dose to the target
volume whilst simultaneously sparing organs at risk is generally the physicist’s
problem (see chapter 2). The physicist must optimize the free variables (such as
the number of fields, the position and shape of the fields etc) to try to achieve this.
Deciding the target dose level, is a clinical decision which must be taken with
adequate knowledge of the biological response of both normal and pathological
tissue (see section 1.1.11). Dobbs (1992) has discussed the need for clarity in
specifying the location within the target volume where the dose level should be
set. A new ICRU report (ICRU 1993), to be published in 1993, will emphasize the
need to clarify whether this is at the point of maximum, minimum, modal, mean
or median dose value.
I .I -9. Dose display and evaluation of plans
1.I .9.1. Methods to display dose distributions; dose-volume histograms.

Traditionally the dose matrix has been displayed as a series of isodose lines. The
reason for this was threefold. Firstly, it emulated the output from ‘hand planning’;
secondly, in the early days of computer planning, grey-scale display of dose as
a series of pixel values was not possible; thirdly, smoothly varying isodose lines
could be interpolated from a smallish number of data points calculated on a coarse
grid. Today the digital 3D dose matrix can be interrogated in many more ways.
Once the 3D dose matrix has been computed, the results may be displayed
superposed on grey-scale anatomical images from CT or MRI. Generally it is

12) 11 1 (31 A
Figure 1.9. Ways of displaying the results of a 3 0 dose calculation: (1)

isodoses are drawn in a plane (here transaxial) and superposed on the
grey-scale image (CT,MRI, SPECT or PET); (2) isodose ribbons wrap around a
3D shaded-su@ace image of the target; (3)axonometric display of transaxial
(T),sagittal (S)and coronal ( C ) images with superposed isodose lines in all
three planes-the images need not befrom the same modality; ( 4 ) differential
dose-volume histogram.
adequate to evaluate the correspondence between isodose contours and target

volumes by visually inspecting selected tomographic planes (figure 1.9). A useful
facility is the display of dose with colour wash on the grey-scale anatomy. If
the colour wash can be independently windowed, it is straightforward to visually
perceive what fraction of a target volume has received dose within specified limits.
If sharp and identifiable transitions in colour are used at discrete doses, these
transitions define isodoses in the colour-wash display. Similar superposition of
dose on anatomy may be performed with shaded surfaces in three dimensions
(e.g. Rosenman 1991, Boesecke et a1 1988, Tepper and Shank 1991). A facility
is needed to view dose on anatomy in sagittal, coronal and oblique as well as
the more usual transaxial slices. Some systems (e.g. UMPLAN) can produce
an ‘axonometric’ display whereby selected slices in different directions (and if
required even from different imaging modalities) can be displayed as a single
perspective view. Isodose lines may be overlayed. Facility to retrieve ‘dose at
a point’ and to display confidence limits are also desirable. Experience seems to
have shown that ‘movie loops’, showing sequential slices of dose and anatomy,
are not very successful (Tepper and Shank 1991).
However, it is in recognition that both the geometry and dose are known
in complete 3D digital form that numerical evaluation of plans can be made
with 3DTP. The cumulative (or integral) ‘dose-volume histogram’ (DVH) can
specify the fraction of the target volume raised to specific dose values. This

Dose Dose
Figure 1.10. Left-a schematic integral dose-volume histogram. The

volume Vf is the fraction of the total volume irradiated to the stated dose
or more. Right-a schematic of the corresponding differential dose-volume
histogram. The volume uf is the fraction of the total volume irradiated to the
stated dose. The left curve is the integral of the right curve. An ‘ideal’integral
dose-volume histogram for a target volume would be a square box showing
all the volume receiving a uniform single-valued dose. The corresponding
differential histogram would be a delta function.
may be computed from a differential dose-volume histogram (figure l.lO)t. This

is, however, one of the most time-consuming tasks computationally and there
is interest in questioning whether the information from coarser grids might be
adequate. The DVH is computed in its integral (cumulative) form from the
differential histogram showing the number of dose voxels raised to each dose
level. Wong and Purdy (1990) argue that the DVH is an effective tool in evaluating
rival 3D plans. Indeed it is, but even single multi-planar 3D plans computed by
optimization techniques such as ‘inverse computed tomography’ (see chapter 2)
can be nicely quantified via the DVH for normal and pathological tissues. Kessler
et a1 (1991b) show how those voxels in a dose distribution corresponding to a
specific part of the dose-volume histogram can be interactively displayed, thus
recovering the spatial information otherwise lost.
Drzymala et a1 (199 1) have summarized some good practices when constructing
and using the DVH:
1. It is convenient to include both absolute and relative dose and volume by
labelling all four sides of the plot.
2. When comparing rival plans, the DVHs should be on the same plot for visual
comparison.
3. Multiple differential histograms on the same plot are much more difficult
to interpret than multiple integral histograms, which are preferred.
4. Histograms can be smoothed by drawing a line through the right-most
comers of each bin.
t The term differential dose-volume histogram is used here to mean (as the mathematics
says) the histogram of fractional volume receiving stated dose. This is to contrast with
the commonly used term ‘integral dose-volume histogram’ meaning the fractional volume
receiving the stated dose or more.. Clearly then the former is the derivative of the latter and
the latter is the integral of the former. It is perhaps unfortunate that the word ‘differential’
has entered the jargon here because it is strictly redundant. It could be wrongly interpreted
that these two histograms were two differentiations (or integrations) apart.

5 . Using a non-linear axis for volume may help observe fine structure in the
curves and aid an interpretation of the prediction of normal tissue complication
probability and tumour control probability from the curves (see section 1.1.11).
Small differences in the shape of the DVH curve can have large effects on these
and predictions should be treated with caution.
In summary, the DVH should never be used as the sole method for assessing a
treatment plan but as an adjunct to visually inspecting dose distributions. Neither
should any single parameter derived from it by a biological model be implicitly
trusted.
If a digital dose prescription is available then a dose-difference matrix can be
computed by subtracting the optimized plan from the prescription. Such a dose
difference can then be displayed in the same ways described above as the dose
matrix itself (e.g. by colour wash on selected slices or shaded-surface anatomy)
(Tepper and Shank 1991).
1.1.9.2. Scoring dose plans by ‘regret’. Shalev et a1 (1987) and Viggars and
Shalev ( 1990b) have addressed the problem of basing a decision on the degree of
acceptability of a treatment plan in terms of simple parameters. They specify the
maximum and minimum dose (dose limits) for the target volume and the ‘tolerance
volume’-the volume of such targets which may go above or below these values,
respectively. Maximum dose to non-target volumes and a corresponding tolerance
volume are also specified. Let be the prescribed tolerance volume for the ith
region, Di the dose limit for this region and V (Di)the actual volume which falls
outside the prescribed limit, known as a ‘volume of regret’. The score function
for the ith constraint is
A score of 10 is ideal; no volume falling outside the constraint; a score below zero
is unacceptable, more volume than tolerable falling outside the constraint. Scores
between 0 and 10 are acceptable. Viggars and Shalev (1990a) and Shalev et a1
(1 99 1a) compare rival plans by observing the score functions as changes are made.
For example conventional plans were compared with conformal plans (Shalev e!
a1 1991b). The interactive system is called OSCAR, standing for ‘objective scoring
with coloured areas of regret’, and implemented on an AECL Theraplan system.
As well as generating numerical scores, OSCAR combines the advantages
of viewing 2D isodose distributions with the advantages of the dose-volume
histogram. From the DVH the spatial regions in which dose falls outside the
acceptable limits (volumes of regret) are colour-shaded to show the spatial
locations where such regret occurs. OSCAR utilizes six colours; mild target
overdose = light orange, severe target overdose = dark orange, mild target
underdose = light blue, severe target underdose = dark blue, non-target tissue
overdose = red, specific organ overdose = purple. This is no! the same as colour
washing grey-scale images. Colour is applied only where there is regret and it

0 " ' "
0 15 30 45 60 75
Dose (Gy)
Figure 1.11. Cumulative dose-area histogramfor a mid-plane target region

in seven patients with adenocarcinoma of the prostate. The inverted triangles
represent tolerance points for target overdose and the upright triangles are
tolerance points for target underdose. Open symbols are mild; closed symbols
are severe tolerances. Note the histogram weaves between these constraints.
This demonstrates a satisfactory treatment plan has been found @om Hahn
et a1 (1990)) (reprinted with permission from Pergamon Press Ltd, Oxford,
UK).
is specifically coded to the type of regret (Viggars and Shalev 1990b). These are
'images of regret'.
It is important to remember that the areas shown by colour-wash are not washed
by dose values. Parts of the washed regions may well be within tolerance. The
wash is applied everywhere to a volume where the overall score function for that
volume lies outside tolerance.
The score function can lead to a different way of organizing treatment planning.
The clinician is asked to provide a prescription, a series of targets and organs at
risk together with the fractional volumes which may or may not be raised to certain
tolerances. Provided the treatment planner derives a scheme with positive score
function, the plan can be considered acceptable and in principle need not be shown
to the clinician-the prescription having been met. Another way of interpreting
this is that if the DVH follows a certain track in space, weaving between limits
set by the clinician, then the plan has to be acceptable. It may not however be
entirely clear how to clinically set the dose limits or how to interpret the regret if
dose limits cannot be met.
This is illustrated in figure 1.11. The clinician requested no more than 20%
of the target area (this was a 2D plan) received dose above 73 Gy and no more
than 50% above 69 Gy. Conversely no more than 5% receives 59 Gy or less (i.e.
no less than 95% should receive 59 Gy or more) and no more than 50% should
receive 63 Gy or less (these are the F and Di in equation (1.1)). These four points
are shown in figure 1.11. The curves represent successful plans (positive score
function) because they weave between the constraint points (Hahn et a1 1990).

Shalev et a1 (199 1b) measured changes in the DVH and in the score functions as
a result of movement at the time of treatment and thus quantified the importance of
movement. They found for example in a prostate study that f0.5cm movement
led to changes in score function of about 5 units.
1.1.IO. Verifying the treatment plan

After treatment planning is completed, the patient may retum to the simulator
where the optimized fields can be set up. The DRR can be checked against the
corresponding physical radiographs. Similar checks can be made at the time of
treatment if imaging is available on the treatment machine. It is apparent that the
overall precision in radiotherapy is compounded of potential errors at all stages
of the process and errors at delivery have possibly received until now too little
attention. One must remember the maxim that ‘a chain is only as strong as its
weakest link’. This subject is considered in depth in chapter 6.
1.1.11, Biological response of normal and pathological tissue

It is at this stage that attention must be turned to folding in the biological
response data to arrive at some statement of the anticipated effectiveness of
treatment. A display of the volumetric distribution of the tumour control
probability (TCP) and the normal tissue complication probability (NTCP) would
be most useful (Wong and h r d y 1990). Until comparatively recently, treatment
planning was confined to establishing acceptable dose distributions, normalized
to some ‘100%’ value which was then assigned to some value in Gy by clinical
decision. Sometimes this has been called the modal dose, but this may be
confusing as it could be interpreted to mean the most frequently occurring dose
value. The basis for deciding this value lies in what is known of the way the
probability of tumour control (TCP) varies with dose and the corresponding way
the normal tissue complication probability (NTCP) varies with dose.
The tumour control probability is the probability of eradicating all tumour cells
plotted as a function of dose. The normal tissue complication probability is the
probability, as a function of dose, of inducing some particular complication (a
collective word for describing a variety of conditions such as nausea, vomiting
etc) in a normal (ie non-tumour-bearing) organ.
1 .I J1.1. Homogeneous irradiation. For radiotherapy to be effective at all it is

obvious that the curve for TCP must lie to the left of the curve for NTCP when
target dose is plotted along the abscissa (see figure 1.12). The closer the curves,
the more difficult it becomes to select a dose level which is likely to control the
tumour without causing normal tissue damage. This is particularly problematic if
the target volume and organs at risk are sufficiently close geometrically that they
receive similar dose levels (for example if the organ at risk lies within a concavity
of the outline of the target volume).

.+ 75
0
Control I Normol
‘: Tissue
e
A b s o r b e d dose
15
Normal
h O
... I
Absorbed dose
Figure 1.12. A schematic showing the tumour control probability and

normal tissue complication probability as afunction of dose. Curvesh show a
favourable situation where a high tumour-control probability can result with
a small complication probability. Curves B show a less favourable situation
where a high tumour-controlprobabilitywould result in a larger complication
probability (from Wambersie et a1 (1988)).
(UT[)
Absorbed dose
Figure 1.13. The ‘mantelpiece clock’ overlap curve of TCP and ( I - M C P ) .

Curves A and B correspond with using the curvesfrom figure I .12. This shows
the probability of uncomplicated tumour control (Pu& (from Wambersie et
a1 (1988)).
The probability of uncomplicated tumour control is simply given by PUTC=

T C P (1 - N T C P ) . If the individual probability curves are sigmoidal in shape,
as they generally are, then the shape of PUTcis like the outline of a mantlepiece
clock (see figure 1.13) (Wambersie 1988). The optimum dose level is sometimes
defined as the dose level which maximizes PUTC.We may see that the closer the
curves for TCP and NTCP, the lower is the probability of uncomplicated tumour
control if target volume and organs at risk receive the same dose. It is again a
clinical decision how to balance the risks. Often the (sic) modal dose is determined
not so much to get a high probability of tumour control but to keep the probability
of normal tissue complications very low indeed (some 1-296) on the basis that
quality of life counts high (see also section 1.1.9). This is referred to as ‘treating
to tolerance doses’ (Emami et a1 1991).

Figure 1.14. A schematic showing the benefit of confonnal versus

conventional therapy. In the left-hand diagram showing conventional therapy
a 90% isodose contour has been shaped to the target volume T, but the
non-conformal technique leaves an organ at risk, such as the spinal canal S,
within the 60% isodose contour. The inclusion of the spinal chord within the
60% contour is an example of ‘regret’.In the right-handdiagram a conformal
technique has reshaped the isodose contours so that the organ at risk is outside
the 60% line. At constant TCP this would lead to a lower NTCP.Alternatively
the dose could be increased to give the same NTCP as in the left-hand example
with a correspondingly higher TCP.
One way round the problem is to design therapy so that the dose to organs at risk
is much lower than to the target volume-the goal of modem confonnal therapy
(figure 1.14). This gets us away from the simple situation where a single vertical
line in the plot of probabilities gives the TCP and the NTCP and thus lessens the
need for the two curves to be wide apart.
1.1 J 1 . 2 . Inhomogeneous irradiation. The above description of how dose

maps to biological effect is too simple. In all treatment plans there is an inevitable
variability of dose to the target volume and a separate variation in dose to the
organs at risk. It makes no sense to speak of the dose level to the target volume
and the dose level to the organs at risk. Rather we should map the distribution
of dose into a distribution of biological effect, dose being simply a ‘surrogate’
for biological effect (Goitein 1991). But here begins the difficulty. Whilst
some data are available (e.g. Lyman 1985, Emami et a1 1991) and methods
of using it (Kutcher and Burman 1989) the data are incomplete and open to
questioning (McShan 1990b). It is not too difficult with state-of-the-art computer
planning systems to generate dose-volume histograms for target volume and
normal tissue; converting these into an integral probability of biological effect
is more controversial (Nahum and Tait 1992). It is sometimes not too difficult
to do the mathematics, but more difficult to accept the predictions of the results.
Let us first look at methods of evaluating the biological effect of inhomogeneous
irradiation which have an intuitive appeal.
The problem is how to convert from a DVH to a measure of risk. In
particular a method to choose between plans which create intersecting dose-
volume histograms is essential to understanding the trade-off between plans which
create small volumes raised to high dose compared with those which create large
volumes raised to lower doses (see e.g. Kutcher et a1 1991). In the literature can

be found curves for the complication probability as a function of dose for different
tissues and complications. These are generally for whole organ irradiation. The
differential dose volume histogram shows the fractional volume of tissue raised
to each dose level. Essentially the differential DVH has to be folded into the risk
curve to compute the cumulative risk. In order to do this properly, account must be
taken of the dose-volume relationship. If (in the differential DVH) the fractional
volume vi receives dose Di, then the equivalent whole-organ (WO) dose (for that
fractional volume) is
OWo = Dj (1.2)
where n is a tissue-specific parameter.
(Tissues with a small value for n are said to have a ‘small volume effect’ and
vice versa. As n tends to zero, the complication probability becomes independent
of volume. Burman et a1 (1991) have fitted the biological data of Emami et a1
(1991) and provided values of the parameter n for 28 normal tissues with defined
endpoints. These values range from n = 0.01 (ear) to n = 0.87 (lung).)
Using this whole-organ dose, it is now allowable to read off the
N T C P (1, DWO) for that partial volume where the 1 implies the risk curve for
the whole organ and OWo is the dose adjusted to the equivalent whole-organ dose
from equation (1.2). The total risk is then (Shalev et a1 1987)
NTCPtot = [1 - l 7 j (1 - N T C P (1, OWo))]. (1.3)
(See, however, equation (1.20): equations (1.2) and (1.3) hold only for small
values of the NTCP. This is not a major limitation because clinical radiotherapy
aims to work at the low end of the NTCP curve.) Support for the power-law
model comes from Kutcher et a1 (1990), who fitted the radiation-response data
of Hopewell et a1 (1987) for white marrow necrosis in the rat but the model’s
general validity is doubted by Yaes (1 990).
Another way to achieve the same end is to assume the risk behaves as
some power law of volume and, from the risk curve for full volume irradiation
N T C P (1, D ) , derive the set of risk curves for partial volumes N T C P ( U , D )
(see figure 1.15 and equation (1.14), where the power-in-volume law is given).
The differential dose-volume histogram can then be folded fractional volume by
fractional volume into the appropriate curve, without the need to convert the dose
Di to an equivalent whole-organ dose, to derive an overall risk, i.e.
Yet another way is to use the whole-organ NTCP curve, but not convert the dose
Di to whole-organ dose OWo, in which case a power of U appears in the equation,
namely
NTCPtot = [1 - Ill [(l - N T C P (1, Di))lUi] (1.5)
(see also equation (1.11)).

Figure 1.15. Schematic showing the relationship between curve ( a )for

NTCP( 1, D ) , giving the normal tissue complication probability for the whole
organ irradiated at uniform dose D , and curve (b)for NTCP(v, D )giving
the normal tissue complication probability for fractional volume v irradiated
at the same unifbrm dose D. As U decreases, curve (b) moves to the right
giving a lower NTCP at any given dose D. Curves ( a ) and (b)are related
through equation (1.14). Also shown is a dotted horizontal line of constant
NTCP. The doses D1 and D, are linked through equation (1.201,the 'iso-effect
dose-volume relationship'. This holds for all (notjust small) NTCP provided
the power law model expressed by equation (I .14) holds.
One way round these apparent computational difficulties is to reduce the

cumulative DVH to an effective dose to the whole volume through one of the
schemes proposed (e.g. Lyman and Wolbarst 1987, 1989). Then one may
genuinely use the risk curve for whole-volume irradiation and read off the risk
at that effective dose. The other popular scheme (Kutcher and Burman 1989) is to
reduce the differential DVH to an effective volume at the maximum dose and use
the appropriate risk curve for that volume. We shall go into more detail in section
1.1 . I 1.4.
So we see in general there is a great deal of difficulty converting from
inhomogeneous irradiation DVHs into cumulative risk. For this reason the use
of score functions, as in section 1.1.9.2, can be a useful rule-of-thumb indicator.
It is also important to realize that clinicians may well be so keen to set a very
low level of risk of normal tissue damage that all plans accepted will in practice
give risks which are below clinical detection. If this is the case, the data to relate
normal tissue risk against dose delivered are never accumulated! In the study of
Shalev et a1 (1987) the predicted risks of rectal stricture were only a few % and
in practice these were not seen indicating that a conservative set of risk data had
been used.
In the study of Shalev et a1 (1991b) it tumed out that the score function was
linear with the risk. Since the score function is linear with volume beyond
tolerance, this means the risk was linear with volume beyond tolerance (at least
for the prostate treatments and at low risk values). It should not be interpreted
that this is always the case. For example, irradiating just a small part of the spinal
chord beyond tolerance will clearly render the whole motor system ineffective.

This is an example of a ‘series organ’ in the language of Brahme et a1 (1992) and

KaIlman et aI (1992a,b).
1 .I.11.3. The dose-versus-biological-effect controversy. Wheldon (1988) has

discussed the philosophy of applying over-rigorousmathematics to biological data
in some depth.
Why are the biological data for converting dose to a probability of damage so
controversial? Three reasons can be identified
1. Populations vary, Not all patients respond the same, but are assumed to
do so. It is convenient to discuss statistical ensembles rather than individuals
(Wheldon 1988). Emami et a1 (1991) have recently gathered together what is
probably the most comprehensive survey so far of normal tissue tolerance to
radiation. They tabulated the doses giving N T C P = 5% and N T C P = 50%
for whole-organ, 2/3-organ and 1/3-organ irradiation for 28 critical tissues. This
establishes just two points on the NTCP curve at each volume from which the dose
variation of NTCP can be fitted. The volume dependence is also established. Each
set of data is accompanied by a detailed discussion of its source. Burman et a1
(1991) fitted the data of Emami et al (1991) and provide very useful tables and
graphs showing the dose and volume dependences of the NTCP.
2. In documenting the effect of radiation on tissues, the dose is often assumed
to be uniform in the target volume or is simply measured at one particular point
(e.g. Hanks’ (1985) ‘pattern of care study’). In practice, the dose varies in the
target volume tissue and so it is hard to establish accurate dose-response data
from in-vivo irradiations.
3. Animal and cellular model extrapolations to humans are risky.
Goitein (1991) has wamed of the maxim, well known in the computer field,
known as ‘garbage in-garbage out’; in other words the predictions of biological
effect are only as good as the data on which they are based and the models in
which these data are used. Cunningham (1983) wrote of the hope that one of the
major contributions of CT to treatment planning might be its ability to generate
accurate dose-response data for specific tissues. In principle, mapping 3D dose to
anatomical CT data could indeed generate curves of NTCP versus dose, provided
an acceptable way were found to reduce the inhomogeneous irradiation to an
equivalent uniform dose. At present these techniques are still controversial. Also
in-vivo NTCP is kept low in clinical practice so the data at high NTCP are not
gathered.
However Goitein (199 1) succinctly sums up why the future of radiotherapy
planning must take account of biological models:
1. They can give a ‘back of the envelope’ estimate of what improvements might
result from a change in dose level or distribution. They can show what effort in
optimizing dose planning is worthwhile.
2. They can help choose between rival plans with different DVH; indeed they
may show that both DVH curves for rival plans might be below the level at which

biological effect would be noticed and in this respect they would show that rival
different dose plans were biologically indistinguishable. If one rival were more
easily achieved practically, this would be useful information to know.
What inaccuracy actually matters in radiotherapy? This is a difficult question
and there is no one answer. The way towards an answer lies in the data for the
relative steepness of the dose response curves for the tissues irradiated, both target
and normal. For normal tissue the relative steepness can be expressed as that
percent change in dose changing the TCP from 50% to 75% and for normal tissue
the steepness is the percent change in dose changing the NTCP from 25% to 50%.
The former vary from 5 to 50% depending on tissue and the latter from 2 to 17%
(Wambersie et a1 1988). In principle such figures provide a means to answer the
question whether geometric positioning errors at the time of treatment are or are
not important. Another way is to determine restrospectively what effects can be
detected clinically. Goitein (1985) suggested computing ‘extreme error plans’
corresponding to the worst-case geometric errors with the patient in worst-case
positions.
Goitein (1982) discussed at length the question of accuracy in dosimetry in
radiotherapy in terms of the accuracy required of CT data used in treatment
planning. Quoting Geise and McCullough (1977), 2% dose uncertainty is
associated with a spatial resolution of 5 mm in the assessment of organ position.
Since all diagnostic CT scanners have much better spatial resolution than this, CT
resolution is not a limitation on dose accuracy. Even the so-called ‘non-diagnostic
CT scanners’ (Webb 1990) achieve a spatial resolution of 5 mm. Patient movement
of the order of 5 mm would however lead to this kind of uncertainty.
It can be easily shown that random errors in attenuation coefficient measure-
ments in CT are not an important practical consideration for the calculation of
dose. On the other hand a systematic error could be very important. 3.5% error
in the CT number would give a dose uncertainty of 2% (Goitein 1982). Another
cause for concern is the proper determination of the linear or bilinear relation-
ship between CT scanner number and electron density, which should be measured
experimentally for each CT scanner (see e.g. Parker er a1 1979, Ten Haken et a1
1991). Other uncertainties in 3D radiation therapy planning have been considered
by Urie et a1 (1 99 1).
1.I J1.4. Formal equations of the biological model. The formal equations
giving the normal tissue complication probability (NTCP) when a partial
(fractional) volume U is irradiated to a uniform dose D have been given by
Lyman (1985). This is sometimes called the integrated normal (or probit) model
(Schultheiss et a1 1983):
N T C P = I/&
with x a free variable and
s_& exp(-x2/2)~ (1.6)
t = [D- DSO(U)] / [mDso(v>l (1.7)

TYPE A
Figure 1.16. Complication probability for liver, displayed as a three-

dimensional surface as a function of dose and partial volume. For this organ
the volume dependence is represented by n = 0.32. D50 = 40 Cy for whole
organ irradiation with liver failure as the end-point. The slope parameter is
m = 0.15. Type-A curves show the complication probability as a function
of dose at constant partial volume. Type-B curves show the complication
probability as a function of partial volume at constant dose. Type-C curves
show the partial volume as a function of dose at constant complication
probability (from Burman et a1 (1991)) (reprinted with permission from
Pergamon Press Ltd, Oxford, UK).
Here D50 is the dose which results in a 50% complication probability for some
specified complication or end-point; D ~ (U o = 1) is the value appropriate to
irradiating all the volume and D50 (v) is the value for partial volume irradiation.
These two are related through a power law (with index n ) of the partial volume
U , this being the fractional volume of the whole organ. m is a slope parameter
which characterizes the shape of the NTCP curve, which is sigmoidal, being the
integration of a Gaussian with standard deviation mD5o (U). For any particular
tissue and specified end-point for normal tissue damage, the values of the
parameters 0 5 0 (v = I), n and m must be determined from clinical observation.
Burman et a1 (199 1) have tabulated these three parameters for 28 tissues whose D5
(the dose which results in a 5 % complication probability) and D ~ wereo tabulated
by Emami et a1 (1991). For a discussion of the difficulties of the fitting procedure,
see Burman et a1 (1991). Some tissues demonstrate an additional threshold
behaviour, there being a partial volume below which no damage appears to occur.
The model expressed by equations (1.ti)-( 1.8) connect complication probability,
dose and partial volume. The interrelation between these can be represented by a
surface, as shown in figure 1.16.

In general the treatment planning will, however, not yield a uniform distribution
of dose either through the target volume or the normal structures (organs at risk),
the fraction of volume raised to each dose level being characterized instead by the
differential or integral dose-volume histogram (DVH). For the Lyman equations
to be easily called into play, a reduction scheme to reduce the DVH to an effective
fractional volume u,e, uniformly irradiated to the maximum dose D,,,, must be
found. One such scheme has been given by Kutcher and Burman (1989):
(1.9)
where volume U; receives dose D; in the differential DVH and D,,, is the
maximum dose to part of the normal tissue under consideration; n has the same
meaning as before. It has been suggested that dose steps of some 1 Gy are adequate
to compute ueg. This effective volume method proceeds from the conjecture that
each volume element of the differential dose-volume histogram independently
obeys the same dose-volume relationship (equations (1.2) and (1.8)) as the whole
organ (Kutcher et a1 1991). Using Dmax as the dose to which all elements are
transformed ensures that ueff is always less than unity. It is the fraction of the whole
organ volume. If the whole organ is not covered in the CT scanning process (and
hence in construction of the DVH), a so-called 'standard' ICRP volume should be
used. The method is somewhat empirical but Kutcher and Burman (1989) showed
it had many desirable features, for example:
1. it reduces to the same complication probability for uniform irradiation as
was used for the input data, and
2. when small hot or cold spots are introduced to an otherwise uniformly
irradiated volume, the NTCP marginally rises or falls respectively, the amplitude
of the change depending on n.
These equations enable the NTCP for inhomogeneous irradiation of any normal
tissue whose biological parameters are known to be calculated from the DVH.
One may write the normal tissue complication probability (equation (1.6)) as
N T C P ( U , D , m , n , D50 (v = 1)) to indicate the full dependences. Using the
reduction method one looks up N T C P ( v , ~ ,D,, o = 1)). If there
m , n, D ~ (U
is more than one such structure the overall N T C Ptotmay be found from
NTCPt,t = 1 - (1 - N T C P l ) (1 - N T C P 2 ) . . . (1.10)
where there are as many producted terms as there are normal structures considered.
For example in the study of Nahum and Tait (1 992) calculating the NTCP for
pelvic irradiation, the rectum and the small bowel were considered to be the
dose-limiting structures of importance and so two terms were used. Schultheiss
et a1 (1987) argue that the aim of clinical radiotherapy should indeed be to
maximize ll [ 1 - w ;N T C Pi], where wi is the morbidity of some complication
whose probability is N T C P ; . This statement is consistent with attempting to

minimize N T C Ptotin equation (1.10) with the addition of the importance of the
separate risks,
The tumour control probability (TCP) is also a sigmoidal curve with increasing
dose and can be determined experimentally or by biological model. Nahum and
Tait (1992) have provided such a model for bladder tumours (see equation (1.32)).
As discussed above, the clinical practice is generally to prescribe a target dose
which keeps the NTCP very small (Emami er a1 1991) rather than optimize the
probability of uncomplicated tumour control, simply because patients can often
come to terms with a life-threatening tumour, whereas their quality of life is
threatened by normal tissue damage as part of treatment. Nahum and Tait’s
study showed that the mean TCP could be improved if (for pelvic irradiation)
all treatments were customized to give an overall NTCP of 5%. However, the
magnitude of the improvement over globally setting the same target dose for all
patients depended critically on the volume parameter n and indicated the potential
dangers of relying on precise mathematical modelling with poorly understood
biological data-as Goitein put it there is great potential for the ‘garbage in-
garbage out’ syndrome.
The NTCP can be deduced from the DVH without reduction. The differential
DVH is the set of fractional volumes U; each with dose D;. Suppose N T C P (1, 0)
is the NTCP when the whole organ is irradiated to dose D, then the NTCP for an
inhomogeneous irradiation specified by this differential DVH is given by
NTCPinhom = 1 - n:=, [ I - N T C P (1, (1.11)

where there are M subvolumes in the differential DVH, each of which may
be considered uniformly irradiated. This equation expressing the dosevolume
relationship as a power-in-volume law was derived by Schultheiss et a1 (1983).
Niemierko and Goitein (1991) have shown that there is actually no need to
compute the DVH in order to obtain NTCPihom. Suppose a normal tissue is
specified by dose Di to N points, then each point can be thought of as representing
a small volume in which the dose is uniform and so it follows that
N T C P ~ o m= 1 - llE, [l - N T C P (1, D j ) ] ’ ” . (1.12)

Niemierko and Goitein (1991) argue that equations of this form for NTCPihom
characterize tissues with ‘critical element architecture’, that is comprising
‘functional subunits (FSU)’,any one of which receiving damage implies damage
to the whole organ function (such as the spinal chord, nerves etc). If Pi is the
probability of damage to the ith FSU then the NTCP for the whole organ is given
by
NTCP = 1 - [ l - Pi]. (1.13)
This equation and those of a similar form in this section give the probability
of damage when tissues have a ‘series’ organization. Kallman et a1 (1992a,b)
give corresponding equations when the structure is partially series and partially
parallel.

From equation (1.1 1) follows the functional form of the volume dependence of
NTCP (see figure 1.15). Suppose dose D is given to fractional volume U only, the
rest of the organ being virtually unirradiated, then the NTCP for the organ is given
by
NTCP (U, D ) = 1 - [I - N T C P ( 1 , D ) ] ” (1.14)
i.e. the N T C P ( U , D ) < N T C P (1, D ) for all fractional volumes and for all
values of N T C P (1, D)?.
If the value of N T C P (1, D ) is very small, equation (1,14) may be expanded
to
N T C P ( U , D ) = u N T C P (1, D ) (1.15)
i.e. for small complication probabilities, the complication probability is a linear
function of the (uniformly) irradiated volume for constant dose. This linearity
obviously does not hold for higher values of the complication probability, when
the full equation (1.14) applies.
An alternative form for the NTCP for full-volume irradiation to that given in
equation (1.6) is
1
NTCP (1,D)= (1.16)
1 (Dso/DIk +
with
k = 1A/m (1.17)
giving the relation between the parameter k and the shape parameter m. k is also
equal to l / n where n is defined by equation (1.8). The arguments have been
attached to NTCP to clarify that this is an NTCP for uniform irradiation to dose D
for the whole organ (v = 1).
This is sometimes called the logistic model and has the advantage of being in
closed form. It differs from the integrated normal model by less than 1%(which is
clinically indistinguishable) with suitable adjustment of parameters (Schultheiss
et al 1983). These authors also made use of the logistic formula together with the
two known doses giving N T C P = 5%and 50%to deduce the parameter k = 1 / n .
t Equation (1.14) is from Schultheiss et a1 (1983) and was derived as follows. Imagine
irradiating a full volume U = 1 at uniform dose D. The probability of escaping injury
is [l - N T C P ( 1 , D ) ] , but it could also be expressed as [l - N T C P ( l / A , D ) l A if A
subvolumes were all irradiated to the same dose D. So
[I - N T C P ( 1 , D)]= [ l - N T C P ( l / A , (1.i)
If only B parts were irradiated (U = B/A), we have
[l - N T C P ( V = B / A , D)]= [l - N T C P ( l / A , (1.ii)
Substituting the RHS of equation (1 .i) into the RHS of equation (1 .ii) gives equation (1.14).
Equation (1.11) is the general form for inhomogeneous irradiation.

This should be contrasted with the fitting method of Burman et a1 (1991) described
earlier.
The logistic formula may be used directly in equations (1.1 l), (1.121, (1.14) and
(1.15). For example, putting equation (1.16) into equation (1.12) we have
N T C Pi*om = 1 - l-IY=l [ 1
k -1/N
+ (Di/D50) ] - (1.18)
Hence the NTCPidom follows directly from a list of N points with their dose
values Di provided the D50 and k for the complication are known.
From equations ( 1.14) and (1.16) we may deduce the dose D, required to be
given to partial volume U to give the same NTCP as dose D1 to the whole organ,
the so-called 'iso-effect dose-volume relationship' (see figure 1.15). We set
NTCP (U, D) = N T C P (1, 01) (1.19)
and substitute from equations (1.14) and (1.16) to get
(1-20)
For very small doses (low complications)
D,, = D ~ v - " ~ . (1.21)
Equation (1.21) is the well known power law between iso-effect dose and
volume (see also equations (1.2) and (1.8)). For example, (with k=10) Do3 =
1.0701, = l.26D1 and D0.05 = 1.3501. However the FSU architecture
model predicts significant departure from this 'law' when the NTCP is high.
Niemierko and Goitein (1991) give experimental data showing better fit to
equation (1.20). Schultheiss et a1 (1983) also present equation (1.20) and discuss
the approximations leading to equation (1.21).
The histogram reduction scheme published by Kutcher and Burman (1989)
(see equation (1.9)) is just one of several analytic reduction schemes proposed.
It essentially reduces the differential DVH for an inhomogeneous distribution of
the dose to an equivalent volume ueff uniformly receiving the maximum dose.
Equations (1.14) and (1.16) then give the NTCP with D set to D, and U set to
veff. A second reduction scheme has been proposed by Lyman and Wolbarst (1987,
1989) which reduces the cumulative DVH to a single efSective dose de^ through
the whole volume such that this would give the same NTCP as the inhomogeneous
irradiation. It does this by starting with the right-most element of the cumulative
DVH, combining the effect of this bin into the bin immediately to the left of it
(figure 1.17). Let us for the moment consider what happens when there are only
two such elements in the cumulative DVH. (When there are more, the process of
combining the right-most two bins simply sequences until there is only one bin
left). A new dose Deff is created intermediate between the last bin at D,and the

LW, and LW,

I
Figure 1.17. A model dose-volume histogram with just two elements for
discussion purposes. The DVH is 'reduced' by the method of Lyman and
Wolbarst (LW) to give an effective dose 0:: to all the volume; and by the
method of Kutcher and Burman ( K B ) to give a fractional volume raised
to the maximum dose Dma.The biological effect of the two are supposedly
VsB
the same (from Niemierko and Goitein (1991)).
second from last bin at 0 2 where the corresponding volumes are VI and V2. We
have
v2 - VI
NTCP (V2, De#) = -N T C P
v2
(V2, D2) + -Vlv2N T C P (V2, 0 1 ) . (1.22)
This equation states that the probability as a function of dose can be linearly
interpolated in volume. The partial volume complication probabilities are
computed from equations (1 -6) to (1.8), which include the volume dependence
of the 0 5 0 . In equation (1.22) upper-case V is used to signify the partial volume
in the cumulative DVH to distinguish from lower-case U used earlier for the partial
volume in the diflerential DVH. When the cumulative DVH has a large number of
elements, equation (1.22) is repeatedly used to combine data from the right until
a single De* results for the total volume.
Niemierko and Goitein (1 991) have very elegantly shown the relationship
between the result in equation (1.22) and the predictions of the critical element
model. Let us apply equation (1.11) of this model to the two-element cumulative
DVH. We get
N T C P i ~ o m= 1 - [l - N T C P (1, D2)Iu2[ l - N T C P ( 1 , Dl)]" (1.23)
where, as before, lower-case U represent fractional volumes in the difSerential

DVH. But
U2 = v2 - v1 (1.24)
and
U1 = VI (1.25)
and so we have

But from equation (1.14) expressing the volume dependence of NTCP at any given
dose (upper- or lower-case V may be used here)
l-NTCP(l,D)=[l-NTCP(V,D)]"". (1.27)
Combining equations (1.26) and (1-27) we have

v -v
NTCPinhom= 1 - [ l - N T C P ( V 2 , D 2 ) ] v [ l - N T C P ( V 2 , 0 l ) J Z .
(1 -28)
Now expanding binomiallyfor small NTCP we have
2 - VI
v
N T C Pinhom = -
v2
N T C P (V2, 0 2 ) + -VIv2N T C P (V2, 01) (1.29)
which is exactly the same as equation (1.22) of Lyman and Wolbarst (1987,1989).
So we have the important result that this Lyman and Wolbarst reduction scheme is
identical to the critical element scheme ofNiemierko and Goiteinfor small normal
tissue complication probabilities. It is not the same otherwise. This identity
arises because both histogram reduction schemes incorporate the power-law dose-
volume dependence. As we have seen this power-law dependence breaks down
in the critical element model for large NTCP and so the methods would in these
circumstances not be expected to give the same results. However since in clinical
practice most dose planning is worked out so that the NTCP is very low, it is
clear that both methods are, with this approximation, the same and could be used
interchangeably. If NTCP is not small one must not make this assumption.
Niemierko and Goitein's critical element model can predict a Deff and Veff like
the Lyman and Wolbarst and Kutcher and Burman terms which in the limit of
small NTCP reduce to their results exactly but which are different for high NTCP.
These results are (see Niemierko and Goitein 1991):
(1.30)
and
(1.31)
Let us work a simple numerical example to show the similarity of all the
methods at low noma1 tissue complication probability. Suppose some single
normal organ at risk receives 30 Gy to 0.4 of its volume and 40 Gy to the other
0.6. (It would be hard to see how such a simple distribution would arise in practice
but this is just to illustrate the point.) From this differential DVH we see that the
cumulative DVH has all of the volume at 30 Gy or more and 0.6 of the volume at 40
Gy. Let us suppose that for this organ 0 5 0 = 50 Gy and n = 0.1 (so k = 10). We
shall use equation (1.16) to deduce the NTCP for the whole organ irradiated to some
dose (the same qualitative conclusions would be obtained if we used equations

(1.6) and (1.7)). From equations (1.2) and (1.3) (the method converting each
partial volume dose to an equivalent whole-body dose) we find N T C P = 0.0628.
The methods using the doses from the differential DVH without converting them
to whole-body equivalents, represented by equations (1.4) and (1.5) (or (1,l l)),
give the same result since one follows from the other through the use of equation
( 1.14) and N T C P = 0.06 16. The Lyman and Wolbarst method (equation (1.22))
gives N T C P = 0.0606. The Kutcher and Burman method (equation (1.9) with
useofequation(1.14))gives N T C P =0.0615.
We must still deal with the question of how to calculate the tumour control
probability (TCP). This might be determined experimentally but in the absence of
experimental data Nahum and Tait (1992) have provided one model. This is based
on the survival of clonogenic cells (Kallman et a1 1992b). We have
K
T C P = ( l / K ) E e x p ( - N $ (i)) (1 -32)
i=l
where
N, ( i ) = No exp (-ai D ) (1.33)
and NO is the initial number of clonogenic cells (taken to be p,V, where pc = lo7
~ m - V,~ is
, the target volume in cc, D is the dose and a is a value taken from a
Gaussian distribution with mean 0.35 and a standard deviation of 0.08. (K is the
number of samples for a! and should be a large number like lo3 or lo4.) Nahum
and Tait (1 992) found this gave a good fit to some biological data (figure 1.18). It
generates a characteristically sigmoidal curve for TCP as a function of dose. The
question still remains of what volume to use and what dose when the DVH for the
target volume exhibits inhomogeneous irradiation. One option is to use the 90%
volume and the dose to which 90% of the volume is raised. This ignores the ‘rind’
of low dose (which would pull down the TCP) and also ignores the larger doses
given to small parts of the volume (which would increase the TCP). This is an open
question.
A potential danger of taking the biological model too seriously is that since
most irradiations are set to give very low NTCP, the conditions for determining
the precise form of the NTCP curve up to values of N T C P = 1 are not
achieved clinically. Also not enough time has yet elapsed for the predictions
of even small NTCP to be checked out with real future damage. There may be
considerable uncertainty in the values of the volume dependence parameter n by
f0.08 (Burman et a1 1991). However, Kutcher et a1 (1991) have shown that rival
dose plans still rank in the same order, as n is varied within its range of uncertainty.
Hence the problem of not knowing n accurately affects absolute calculations of
NTCP rather than relative calculations. For this reason Sontag and Purdy (1 991)
and Webb (1992) felt confident in using TCP and NTCP to rank treatment plans.
Also there may not be agreement about how to weight the importance of different
complications. It is all very well to know the NTCP but how should one decide the

40 50 60 70 80 90 100 110 120

DOSE (Gy)
Figure 1.18. Showing the tumour-control probability as a function of dose

based on the clonogenic cell-kill model with ct = 0.35,pc = IO7 and
VT = 320 cc with ct ( a ) not randomized and (b)randomized (from Nahum
and Tait (1992)).
importance of a complication independent of its probability of occurrence? Until

these problems are dealt with some workers will prefer to stay with considerations
of dose optimization rather than optimization of biological outcome.
On the other hand, there are several practical advantages accruing from the use
of biological models:
1. They can give simple numerical data (the NTCP values) which help answer
the question ‘which is the better of two cumulative DVH’s which cross-over?’;
2. They reduce the staggering amount of data in treatment plans that have to
be analysed to manageable proportions; and
3. They provide a measure on a biological rather than a dosimetric scale.
1.1J1.5. The time factor and inhomogeneous irradiation. Time also enters the
consideration of biological effect. This is a big subject, well treated by others (e.g.
Wheldon 1988) and here we are brief. So-called ‘iso-effect formulae’ predict
the way in which the dose per fraction, the fraction size and the total dose may
be manipulated to give the same recognizable biological effect. There are many
ways of providing a mathematical treatment. One is to compute a time-dose factor
(TDF) given by
T D F = k , N d s (TIN)-‘ U@ (1.34)
where N is the number of fractions, d is the dose per fraction, ( T I N ) is the time
in days between fractions, U is the partial volume of tissue irradiated (a fraction of
the reference volume U = l), kl is a scaling factor making T D F = 100 for the
reference volume of tissue irradiated to tolerance and 8 , t and 9 are tissue-specific
exponents.

However, once again this oversimple equation ignores the inhomogeneity of

dose throughout a three-dimensional target volume. Orton (1988) has proposed
that target volumes might be considered broken into subvolumes uj in which the
dose per fraction di is uniform. Then because of the power law, the individually
calcuiated (TDF,)”” are linearly additive and one can write an integral TDF
factor (ITDF) as
ITDF =
where
T D F j = klNd! (TIN)-’ U?. (1.36)
A second way to include time is the use of a linear quadratic factor (LQF) given
by
L Q F = kl N d (1 +d/(a/B)) U@ (1 -37)
where the terms have the same meaning as for TDF and ( a / @is ) the well known
ratio for the survival curve (Wheldon 1988). High values of the ratio indicate
tissues relatively insensitive to changes in fraction size and vice-versa.
As for TDF, account for dose inhomogeneity through the target volume can be
made by defining an integral LQF (ILQF) via
where
L
ZLQF = C ( L Q F i ) ” @
I’ (1.38)
LQFi = klNdi (1 + d j / ( a / B ) ) (TIN)-‘ vf. (1.39)

Orton ( 1988) tabulates the tissue-specific factors for skin, stroma, brain, spinal
chord, lung, gut and kidney.
1.2. REGISTRATION OF TWO IMAGE DATASETS FOR 3D TREATMENT

PLANNING
In recent years the proliferation of medical imaging modalities has had a dramatic
effect on the expectations of performing good radiotherapy. An immediate
consequence has been the need to develop techniques for image registration (Chen
et a1 1990). The use of medical images to assist the radiotherapy treatment
planning process is so important that the second part of this chapter develops the
subject of image registration in some detail. In the early days of using images
in radiation therapy planning it was easier and was traditional to merge data on
sheets of paper. For example although CT data might be accessed by computer
for treatment planning, the same software might not be able to accept SPECT
or PET data. Goitein (1982) argues that in these early days the computer was

Registration of two image datasets for 3D treatment planning 39
Figure 1.19. Left-MRI image of a patient with a clival chordoma. The

treatment volume has been outlined by the therapist. Right-the 0 image of
the patient with outlinesfrom CT only and CT+MRI (fromKessler et a1 ( I 991))
(reprinted with permission from Pergamon Press Ltd, Oxford, UK).
a limitation in using multi-modality images for treatment planning. In early times

anatomical structures were translated from the hard copy of one imaging modality
to another by pencil (Sontag and Purdy 1991). These practices are fortunately fast
disappearing.
Imaging methods may be classified in a number of ways; e.g. those
0 generating anatomical data: planar x-radiographs, CT, MRI, ultrasound,
0 generating functional data: MRI, SPECT, PET,

0 generating tomographic volume datasets: SPECT, PET, MRI, CT,
0 with fine scale of resolution: MRI, CT, x-radiographs, some PET,

0 with coarse scale of resolution: SPECT, some PET.
The need to register data arises for several reasons. Sometimes anatomical
detail is more easily seen in MRI than in CT (e.g. differentiating tumour and
oedema from grey and white matter in the brain. Bony structure is best visualized
by CT whereas soft-tissue pathology and nervous tissue at the base of the skull is
best seen by MRI (Hill et al 1991a)). Sometimes the target volume may be defined
differently in two modalities. Correlation of MRI with CT is needed for isodose
calculationswhich require knowledge of the tissue electron densities. This creates
the problem of registering two different modalities but both with high spatial
resolution (Chen et a1 1985,Hill et a1 1991a) (figure 1.19). Additionallyone might
like to make use of functional data, such as SPECT or PET, to determine the target
volume and once again correlation with CT is needed to obtain the necessary tissue
electron densities.
Serial CT or MRI scanning is common, for example pre- and post-operatively;
prior to, and after, radiation therapy. The clinician needs to assess response to
therapy. Hence there is a need to correlate high-resolution data from the same
modality at two or more different scanning times.
Functional data from SPECT and PET is widely available. Radiation treatment

planning is assisted by image combination; e.g. functional PET data may allow
the treatment planner to avoid healthy brain tissue when planning, using CT data,
to treat a brain tumour. The need to match functional data on to corresponding
anatomical images (CT or MRI) creates the problem of correlating two modalities,
one with a high and the other with a lower spatial resolution. The mismatch can
be by a factor 10 (SPECT with CT say)(see e.g. Roeske et a1 1991). Also the need
arises to display high-resolution isodose distributions superposed on to (possibly
low-resolution) functional images when monitoring response.
Used alongside anatomical data, diagnosis may be illuminated by functional
imaging (Bidaut 1991). Conversely the anatomical data can provide the basis for
photon attenuation corrections in forming the functional images. Yu and Chen
(1991) have shown how this improves contrast. Chen et a1 (1991) used the ‘hat-
head’ method (see section 1.2.7) to register MRI and CT data with PET data to
provide anatomical boundary information which improved image reconstruction.
Correlation of patient position at the time of imaging for planning, derived
from CT data, and, at the time of treatment, from planar radiographs (either port
films or electronic portal images-see chapter 6), is also important, for example
to verify that positioning at the time of treatment is the same as at the planning
stage. Registration of digital radiographs artificially generated from the CT data
with the planar radiograph at the time of treatment is required. Comparison of
simulator films with portal images at the time of therapy also requires registration
techniques. Megavoltage portal images (digital or film) can be correlated with
simulator images or digital projection radiographs from CT data and studies using
such correlations are underway (Swindell and Gildersleve 1991).
Matching different modalities is complicated because each has its own
geometrical description with different pixel sizes, slice thicknesses, data
orientations, magnifications and possibly non-aligned longitudinal axes (tilted
transaxial planes). All these problems have to be addressed by image-correlation
methods.
The choice of technique for correlating images has much to do with whether
the intention to do so exists a priori before either scanning session takes place, in
which case external markers may be used, or whether the images to be correlated
already exist, maybe from different centres or departments. The ‘strongest
method’ of ensuring precise image registration is the use of the stereotactic frame.
Some types of frame are however somewhat invasive and they are generally
reserved for head imaging only where precision is most important. Stereotactic
image registration is discussed in a separate chapter along with stereotactic
radiotherapy and radiosurgery. A posteriori image registration obviously cannot
use applied fiducials-marks which appear on both sets of images. If both sets of
data have high resolution, skilled observers may be able to identify corresponding
anatomical landmarks which may then take the place of extemal markers (Hill et a1
1991a). For example bony structure, relative to which the target volume may not
move, can serve this purpose. If however one set of data has much lower spatial
resolution, then this kind of a posteriori identification is generally not possible.

se! 1
Figure 1.20. Two 3 0 imaging modalities have been used to take a series
of tomographic sections through a brain tumour ‘T’ but the slices are not
parallel to each other. Imagine the most general situation where they are
related by an unknown rotation and translation in three dimensions. The
transformation can be determined by the use of external fiducial markers
which appear in both sets of images. Markers a,b and c appear in slices 2,
18 and 13 in set I , respectively, and in slices 6,18 and 9 in set 2, respectively.
By measuring their coordinates in each set, the transformation m a y be found.
The only solution is to arrange attached extemal landmarks a priori. These two
methods (use of extemal fiducials and correlative anatomical recognition) are
collectively referred to as ‘landmark-based image registration’ (figure 1.20). In
the absence of these, one particular a posteriori method may still work for the
head-the so-called ‘hat-head fitting method’ or surface-based registration-see
section 1.2.7.
Image correlation techniques can be divided into 2D and 3D. What follows is
a general 3D description which can easily be collapsed to the corresponding 2D
method.
1.2.1. 3-vector description

Let us assume that two 3D digital image datasets p and p’ are related to each
other simply by a 3D rotation, 3D translation and a scale change. For the moment
disregard any warping or elastic transformations. Each dataset comprises points
assumed to be on 3D rectangular Cartesian coordinates. A representative point in
each set shall be rand r’ which may be represented by 3-element column vectors,
i.e.
(1.40)
and
(1.41)

Then
r’=SRr+b (1.42)
where R is a (3 x 3) rotation matrix, S is a diagonal (3 x 3) scaling matrix and b

represents the translation vector. Equation (1.42) implicitly operates for all points
in the sets p and p’. The rotation matrix R can be broken down as the product of
three separate rotations. Let Rx (6,) be the rotation by 0, about the x-axis, R, (e,)
be the rotation by Oy about the y-axis and R, (e,) be the rotation by 0, about the
z-axis. Then (and implicitly defining an order of sequential rotations)
We can write
1 0
0 COS^, sine,
0 -sine, COS^,
cose, 0 sine,
-sin@, 0 cosey
cos0, sine, Cl)
0 1
The scale matrix in general may be written
0 0
s=(: sy 0 ) ( 1.45)
0 0 s,
where s., ,sy and s, are the scaling factors in the x, y, z directions. If the scaling is
isotropic, S collapses to SIwhere Iis the identity matrix and s is a scalar constant.
Combining equations (1.42), (1.43), (1.44) and (1.43, we have
-s, S X c,
(1.46)
where Sx,y,r= sin8,,,q, and CX,,.,= cosB,,,,,.
Note the rotations and scaling are not commutative. They must be performed
in the order written.

I .2.2. 4-vector description

It is rather clumsy to have the rotation and the scaling as a matrix operation and
yet the translation as an additional (rather than multiplicative) vector. This can be
overcome by defining r and r' to be 4-element column vectors:
.=(i) (1-47)
and
(1.48)
We now write
R, (4)=
1 0
0 cose,
0 -sine,
0
sine,
cos8,
:)
0
0 0 0 1
(1.49)
The scale matrix becomes

0 0 0
.=i;:
and a translation matrix is written as
o o s , o
0 0
sy 0O O1 )
(1S O )
(1.51)
If the scaling parameters sx , sy, s, can take negative values, this will also take care
of the 'handedness' of the datasets, i.e. the possibility that the planes are numbered
in reverse order in one set from another.

Now r and r’ are linked by
(e,) R, (e,) R, (e,) r.

r’ = T S R, (1-52)
It is readily apparent that the (4 x 4) operator R, (e,) R, (e,) R, (e,) acting on

r, as defined by equation (1.47) rather than (1.40), performs the 3D rotation as
in (1.42) but leaves a 1 as the fourth element in the rotated vector (see equation
(1 3 3 ) ) . The left-most operator T S is
TS=(’ ‘Y0 0 ’Y)

b,
(1.53)
0 0 s, b,
0 0 0 1
When this operates on the rotated 4-vector it scales and translates the data leaving a
4-vector J in which the fourth element is unity (see equation (1.57)). To appreciate
this better, create the operation in equation (1.52) in stages: first
(1 -54)
then
x cos 0, - y sin 8, sin 0, +

z sin 8, cos 0,
y cos 0, z sin 6, + (13 )
--x sin e, - y cos e, sin e, z cos e, cos e,
=
R~ X
‘
+
1
and
(e,) R, (6x1 r
R, (6,) R,
xc,c, - yc,s,s, + zc,s,c, + ys,c, + zs,s,
=i
where S.r,y.z=
-xs,c, + yS,S,S, - zs,syc,
- x s , - yc,s,
1
+ zc,c,
and Cx,y.z= cost9x,y,zand finally
+ yczc, + ZCZS,
(1.56)
r’ = T S R, (e,) R, (e,) R, (e,) r =

a4
where

a1 = s,x cos 6, cos e, + s, y (sin 6, cos e, - cos e, sin e, sin e,)

+S,Z (COS e, sin e, cos e, + sin e, sin e,) + b,
a2 = -s,x sin 6, cos e, + s, y (sin e, sin 6, sin e, + cos e, cos e,)

+sYz (cos 0, sin 0, - sin 6, sin 0, cos 0,) + b,
a3 = -s,x sin 0, - s,y cos sin 0, + S,Z cos 8, cos 6, + b,
Qy
a4 = 1. (1.57)
We see that equation (1.57) is (of course) identical to (1.46). The advantage of
using equation (1.52) or (1.57) is that all geometrical operators are represented as
matrix operations. The vectors require to be 4-vectors and the R, T, S matrices
are square (4 x 4) operators.
In order to register two datasets, four or more corresponding points must
be identified in both sets. From these may be determined the parameters
e.,, e,, e,, s, sy, s,, b,, b y ,b, using equation (1.57).
1.2.3. Decoupling translation from rotation and scaling using centre-of mass
coordinates
The way forward to solving for the geometric operators is to uncouple the
translation from the rotation and scaling. This is done by working in centre-of-
mass coordinates. We shall regard the rotation as taking place about the centre-of-
mass of the reference points in thep set. Supposep,, is the centre-of-mass vector
of the set of reference points in p and p&, is the centre-of-mass vector of the set
of reference points in p’. This does not mean the centre of mass of the real tissue
enclosed by such points; instead imagine unit mass at each of the points with air
between. That is
N
Pcm = ( NL ) . ri (1.58)
r=l
and
(1 -59)
where there are N reference points in each of the datasets to be registered. Then
rcm = r -P m (1.60)
and
(1.61)
I I I
‘cm
= r -Pcm*
For the moment return to the 3-vector description. Putting equations (1.60) and
(1.61) into (1.42) we have
r& +P:m = S R (rcm + p c m > +b

(1.62)
The last two terms in equation (1.63) cancel leaving
rbm= S R rem. (1.64)
The image datasets in the centre-of-mass coordinates are now linked only by the
rotation and scaling operators S R.
1.2.4. Methods of solution

There is no universally acceptable method of solving the registration problem.
Kessler (1989) has classified the available techniques into four groups. These
are point-matching, line-matching, surface matching and interactive matching.
The first three of these attempt to force one dataset to coincide with another by
minimizing some merit or cost function representing the mismatch. The fourth
method instead relies on a user attempting to register anatomical (e.g. bony)
landmarks. All methods have the advantage that the imaging volumes do not
have to both encapsulate the entire volume of interest, unlike other methods (e.g.
Gamboa Aldeco et a1 1987) which use moments of the distributions and can only
assume isotropic scaling.
1.2.5. Solving by point-matching method

The point-matching method, as its name suggests, relies on the use of some
external fiducials. These may be small spots of olive oil or CuS04-doped water for
MRI studies, solder for x-ray CT or "Tc* for SPECT or 68Gafor PET. Alternatively,
the points may be observer-determined anatomic point landmarks (e.g. Hill er a1
199la).
1.2.5.1. Solving for the rotation operator. Equation (1.64) describes the
relationship between any of the reference points, expressed in the centre-of-mass
frame, from the two datasets. The operator S R is determined by minimizing the
sum of the squares
N
x2= llri - S R rill2. (1.65)
i=l
where the sum is taken over all the N reference points and implicitly over all the
three indices of the vectors. (Important note: the cm subscript has been omitted for
clarity in the following mathematical development. In section 1.2.5.1 all vectors
are in the centre-of-mass frame.) This method has been used by Hawkes et a1
(1990) and Hill et a1 (1990) for aligning MRI and SPECT data. Their solution

generally regards the scaling as isotropic so only four quantities are calculated
e,, ey,0, and s. Similar 'point matching' has been performed by Kessler (1989)
using from 5 to 12 points to improve the sensitivity of the fit. Chen et a1 (1985)
used point structures visible in both modalities to register MR and CT images for
radiotherapy planning. Evans et a1 (1989) registered PET images with MR images
the same way. Phillips et a1 (1991) registered MRI and CT this way for treatment
planning and Hill et a1 ( 1991a) registered MRI and CT data by a point-feature based
technique. Rizzo et a1 (1990) registered 2D slices from PET and SPECT with MRI
and CT after these had been identified from 3D datasets. Hill et a1 (1990) registered
2D gamma-camera images and x-radiographs this way and also used the method
for 3D registration of MRI and SPECT data.
1.2.5.2. The method offinding the rotation matrix R. First regard the centre-
of-mass dataset r( as also being de-scaled by the operator s, in which case
(1.66)
The method of minimizing x2 in equation (1.66) has been described by Arun

et a1 (1987). This is sometimes known as a 'Procrustes problem' (Schonemann
1966). Expand the right-hand side to obtain, ' representing matrix transpose,
(1.67)
i=l
so
N
x2 = (rl'ri + rf R'R rj - rj'R ri - r;R'ri) , (1.68)

i=l
Because R is a rotation matrix R'R = 1. The last two terms in the above expression
are transposes of each other and identical because these two terms are also the (sum
over i of) the inner products of two vectors r; and R ri (recall the inner product of
two vectors a and b is a'b = b'a and is a simple scalar quantity (Kom and Kom
1968)). Hence these two terms are the same and we can write
(1.69)
We require to minimize x2, which is achieved by maximizing
N
F = Cr,!Rri (1.70)
i=l

This is the trace of the (sum over i of the) inner product of two vectors r; and
R ri. Using the reversibility property of the trace of two matrices (i.e. Tr M L =
Tr L M), equation (1.70) can be written
(1.71)
where the matrix H is defined as

N
H = Erir:. (1.72)
i=l
The problem has now reduced to finding the R which maximizes F in equation
(1.70). A neat way exists to achieve this. First it is necessary to find the singular
value decomposition of H. The SVD is given by
H = UAVt (1.73)
where U and V are ( 3 x 3) orthonormal matrices and A is a diagonal (3 x 3 )

matrix with non-negative elements (recall that an orthonormal matrix L has L L' =
L'L = I). A method for obtaining the singular value decomposition has been
described by Press et a1 (1 988).
Now we construct the matrix
X=VU' (1.74)
which itself is orthonormal. We can now show that X is the optimum R rotation
required to minimize x2 above. To do this construct
X H = V U'UAV (1-75)
from which
X H = VAV' (1.76)
But this is a symmetrical (because it is equal to its transpose) and positive definite
matrix. It is a theorem from the Schwartz inequality (Gellert er a1 1977-see Arun
et a1 1987) that for such matrices and for any ( 3 x 3) orthonormal matrix L one
cares to choose,
Tr (XH) 2 Tr (LXH) (1.77)
and so of all the orthonormal matrices, X maximizes F in equation (1.70). If the
+
determinant of X is 1, then this is a rotation, which is what we want. X is the
optimum R. If the determinant of X is -1, then this is a reflection, which is not
what we want. The correct X (or R) is formed by changing the sign of the third
column of V corresponding to the smallest singular value of H.

So the algorithm is summarized by equations (1.72)-( 1.74), from which R may

be determined. This was the method used by Arun et a1 (1987), Hawkes et a1
(1990) and Hill et a1 (1990,1991a). Similar methods have been developed by
Evans et a1 (1989). How successful the solution is depends on the noise in the
problem and the degeneracy in the choice of points to register. These problems
are discussed by Arun et a1 (1987) and by Hill et a1 (1991a).
1.2.5.3. Solving for the translation operator. The translation is simply given
by
b = p’, - S Rp,,. (1.78)
Now we have the six parameters required to describe all the solid geometrical
registration, namely e,, e,, e,, b,, by and b,. Whilst these have been determined
using the 3-vector description, it is to the 4-vector description that we return to
implement the results and register datasets by the simple matrix operations in
equation (1S2).
I .2.6. Solving by line-matching method

Kessler (1989) and Kessler et a1 (1991) have shown how to use line-matching
methods to obtain the transformation parameters under circumstances where it is
difficult to ensure that single points are in the field of view. A custom mask for
the head was constructed to which four glass rods were attached; the orientations
of the rods were chosen to maximize the sensitivity of the method. The rods were
filled with the same materials as for point-matching. A merit function involving
the centres of the rods was minimized.
I .2.7. Solving by surface-matching method

This is perhaps the most difficult method to engineer. It requires that two surfaces
can be constructed from each dataset and matched. The obvious application is
for the head, where the skull surface has been used. The jargon ‘hat-head’ fitting
or ‘computer docking’ has come into use to describe this technique (Kessler et
a1 1991b, Kessler 1987, 1989, Pelizzari et a1 1987,1989, Levin 1990, Levin et a1
1988, 1989,1991, Chen et a1 1990, Salehpour et a1 1991, Neiw et a1 1991, Bidaut
1991).
The technique first requires that from each dataset a tiled surface is constructed.
Many methods for doing this have been proposed (e.g. Christiansen and
Stephenson 1985, Fuchs et a1 1990). The vertices of triangular tiles in the rotated
‘hat’ ri are joined to a common point below the ‘head’, The points 6 are defined
as the intersections of these lines with the ‘head’ (figure 1.21). The same merit
function as in equation (1.65) is then minimized. This oversimple description
masks a wealth of computational arrangements.
For example, Neiw et a1 (1991) have recently automated the technique with
the following improvements. The contours from each tomographic slice were

Figure 1.21. The ‘hat-head’ method of image registration. One tomo-

graphic dataset defines the contours of the ‘hat’.Selectedpoints ri are located
along the hat contours. A second tomographic dataset (generally a different
modality) defines the ‘head. All the points ri are joined to a common point
below the ‘head intersecting the head surface at 4. The ‘hat’ is fitted to the
‘head’ by a minimization process.
extracted automatically (by thresholding CT and MR data and by the Marr-

Hildreth (1980) edge detector for PET data). This automation of the feature
extraction phase overcomes one of the major difficulties, the need for human
interaction. The merit or cost function was defined using data which could be
pre-stored. The minimization used simulated annealing to avoid the optimization
becoming trapped in a local minimum.
Levin et a1 (1991) have used this technology for merging images for a variety
of applications, not all connected with radiotherapy. These include:
displaying the surface of the brain in 3D to show distortions of the sulci, gyri
and convolutions resulting from tumours beneath the surface. This can assist
the surgeon operating without damaging the important functions of the residual
brain.
displaying internal tumours through the semi-transparently rendered brain
surface.
developing the technique of computer-simulated craniotomy (using 3D MR
data).
displaying integrated PET and MRI 3D data to show abnormal brain function
correlated with anatomical structures (figure 1.22). When abnormal function
is matched to apparently normal structural details, these merged images can
be used in radiotherapy treatment planning. Levin er a1 (1991) have a ‘plane
picker’ tool. Planes are identified on the 3D displays and the corresponding
2D tomographic sections are displayed. Provided the correct registration has

Registration of two image datasets f o r 3D treatment planning 51
Figure 1.22. Integrated 3 0 display of the brain of a child with intractable

seizures due to extensive encephalitis. There is a mild abnormality of the
lower segments of the motor and sensory strips just posterior to the speech
area. This abnormality, which was confirmed at surgery, was not clearly
shown by the original cross-sectional images Cfrom Levin et a1 (1991))
(reprinted with permissionfrom ‘3D imaging in medicine’,@CRC Press Inc,
Boca Raton, Florida, USA).
been made, these will be exactly corresponding planes from CT, MRI or PET.
Salehpour et a1 (1991) extended the method to make use of multiple matched
surfaces including, for example, the inner table and the frontal sinuses.
Chen et a1 (199 1) incorporated the structural information from MRI and CT
images into PET image reconstruction. To do this they located the important
anatomical boundaries from the MRI and CT data and then showed that when these
were used in a Bayesian image reconstruction scheme (rather than the convolution
and back-projection method), the reconstructed PET images were sharper, as well
as less noisy. This finding was from simulated data and only stated in qualitative
terms.
Bidaut (199 1) implemented the method to assist interpretation of brain images
from PET as follows. MRI and PET data were contoured by a Laplacian of Gaussian
filter, followed by interactive editing. The MRI data formed the ‘head’ and the PET
data the ‘hat’. The matching algorithm minimized the volume between the two.
After registration,MRI data (being the modality with the higher spatial resolution)
were re-sliced to match PET reference slices. Regions of interest were transferred
between the two. Side-by-side display and colour-wash were available. The
implementation was set up in a local framework which allowed for expansion
to include other modalities in the future (e.g. SQUID magnetoencephalography
(Reichenberger 1990)).
Hill et a1 (1991a) point out that this method is only suitable when the skin
surface is itself not distorted by the method of supporting the head and when
there is sufficientarea of skin to provide a good matching. Because of the above
constraints they did not use the method for registering 3D images of the skull base,

instead preferring landmark-based correlation.
1.2.8. Interactive and optimized iterative matching

Guesses for the matching are tried until obvious anatomical landmarks from one
dataset match those in the other. As well as being fast (Evans 1992a,b) this can
also provide the first estimate for the other three techniques. Interactive image
matching is further discussed in chapter 6. This ‘forward’ method has to be used
if the scaling factors are unknown because in those circumstances the analytic
solution of Arun et a1 (1987) cannot be constructed.
Ende et a1 (1992) reported a method of anatomical landmark-based correlation
with the following philosophy. At least three matching anatomical ‘points’ were
selected from each of two unregistered 3D datasets. The points were expressed
as 4-vectors. The two matrix operations taking each set of three points into a
common coordinate system with one point at the origin and a second lying along
a z-axis was determined. From this the single matrix operation taking one set
of three points into the other was determined. By definition, one of the points
from each dataset then has to coincide with its matching relative in the other
dataset. The other two points will in general not exactly coincide. The error in
the transformation was characterized in terms of the mismatch of these other two
points. Two sets of three points can be so matched in six separate ways. By
using more ( n )points the number of possible ways increases to 6n!/3!(n-3)!.The
optimum transformation was selected as that which gave the minimum mismatch
between the transformed sets of (n-1) points. Ende et a1 (1992) found the error
could be reduced to within the spatial resolution of the datasets. They matched
PET, MRI and CT data.
1.2.9. Lock-and-key matching

The registration methods discussed so far have relied on identifying corresponding
points in images from two different modalities, be they anatomical landmarks
identified a posteriori or a priori placed external markers. Hill et a1 (1991b)
addressed the problem of how to register two datasets when corresponding features
were nor able to be imaged by the two modalities. They cited the example of trying
to discover the geometric relationship between major blood vessels on the surface
of the brain and the surface fissures of the brain. The best way to image the blood
vessels is by digital subtraction angiography. The Guys group have developed
methods for reconstructing the 3D blood-vessel structure from biplane angiograms
(called SARA for ‘system for angiographic reconstruction and analysis’). Such
vessels do not image well with MRI, which conversely can easily produce images
from which the 3D brain surface can be reconstructed. Hill et a1 (1991b) used
TI- weighted spin-echo images and developed a manual method of reconstructing
surfaces. The task was then to discover the best solid-body transformation to
locate specific major vessels into the convolutions of the brain surface. Clearly

Registration of two image datasetsfor 3D treatment planning 53
the problem was not to register corresponding structures but to arrange the correct
relationship between two structures (the vessels and the brain surface) known to
be adjacent. To do this, Hill et a1 (1991b) defined and minimized appropriate cost
functions, assumed geometric scaling factors were known and solved for the three
translation and three rotation variables. Several optimization methods were tried
including the simplex and simulated annealing (see also chapter 2). The aim was
to fit two structures together by analogy with fitting a key into a lock. They tested
the method with images of a brain from a cadaver.
The clinical objective was to improve the success of stereotactic procedures
such as biopsy, implantation of radiotherapy sources and placement of sub-dural
EEG electrodes. There is otherwise significant morbidity and mortality associated
with neurosurgical procedures resulting from the accidental damage to blood
vessels. Another application was to combine blood vessel images with tissue-
perfusion images to assess the clinical significance of blood-vessel disease.
1.2.10. Structure mapping

Structure mapping may be performed in two ways. Either:
regions of interest (ROI) and volumes of interest (VOI) from one dataset may be
transformed and displayed on the other, generally necessitating resampling (e.g.
Levin 1990, Chen et a1 1985)
or:
one whole dataset is transformed to match the other. Then a number of
possibilities arise for (i) merging modalities by substituting some of the voxels
from one study by those of another. Hill et a1 (1991a) used this method to
generate very interesting ‘fused’ images of the skull base, wherein the bony
structure was displayed from CT data and all other structure from MRI data (figure
1.23), (ii) side-by-side display (see e.g. Levin 1990), (iii) ‘chequerboard’display
(i.e. ‘white’ squares might receive full colour representing function (SPECTRET),
‘black’ squares coding grey-scale anatomy (MRI/CT) (Hawkes et a1 1990), (iv)
‘modulated display’ (i.e. each pixel colour coded for function but intensity
modulated for anatomy) (Hill et a1 1990), (v) ‘linked cursor’ display, whereby
two corresponding slices are displayed side by side with a line joining equivalent
pixels (Hill et a1 1990, 1991a), (vi) resampled ROIsNOIs, whereby ROIsNOIs
created from one modality are resampled by intersecting planes from the other
(Chen et a1 1985, Kessler 1989).
All the techniques described in section 1.2 assume global inelastic transforma-
tions. There is a large literature covering elastic transformations. Most is beyond
the present scope of this discussion. A good starting place is with Bajcsy and Ko-
vacic (1989), Dann et a1 (1989), Broit (1981), Bajcsy et a1 (1988), Evans et a1
(1991)andGeeetal(1991). HolupkaandKooy (1991) andHolupkaetal(1991)
have also described non-marker-based image correlation.

Figure 1.23. Single fused display of MR and CT data for the skull base. The
bone information isfrom the CT image and the soft-tissue information isfrom
the MR image. The two datasets have been registered in 3 0 by a point-feature
based technique due to Arun et a1 (1987). Images like these were used by the
group at Guy’sHospitalfor surgical planning but could equally wellform the
basis Of 30 radiotherapy treatment planning (from Hill et a1 (1991a)).
1.3. SUMMARY AND THE NCI STUDY OF 3D RADIATION THERAPY

PLANNING
Three-dimensional radiation treatment planning is not synonymous with

planning conformal therapy but it is an essential component in the process.
Starting around 1984, four centres in the US collaborated under contracts from
the National Cancer Institute to evaluate the state of the art. The four Institutes
and their Principal Investigators were: Massachusetts General Hospital (M
Goitein), Memorial Sloan-KetteringCancer Centre (J S Laughlin), University of
Pennsylvania (P Bloch and M Sontag) and Mallinckrodt Institute of Radiation
(J Purdy). Their work took many years to complete and the findings occupied
a whole issue of the International Joumal of Radiation Oncology, Biology and
Physics (May 1991) running into several hundred pages. This report is a seminal
publication in this area. Many of the technologies discussed have been reviewed in
this chapter, but readers should consult the lengthy papers for detailed discussions
of the collaborators’ findings. The NCI Contractors studied just 16 patients
(275 treatment plans) in a common framework in order to make meaningful
comparisons. To conclude this chapter some significant statements are extracted
from these reports.
Any review of the ‘state of the art’ has to decide whether this is to be ‘that
which is commonly available’ or ‘the best which is currently available’ (Sontag

References . 55
and Purdy 1991). The NCI Contractors adopted the latter definition, yet even
within these four Centres it was clear that different technological capabilities
existed (Laughlin et a1 1991). In this chapter, reviewing the work at these and other
Centres, the same approach was taken. The tables in the publications summarizing
the technological capabilities at these four Centres might be regarded as ‘wish
lists’ against which developments elsewhere could be checked.
Certain ‘growth areas’ may be identified including the increased use of tools
for objective scoring of treatment plans, tools for computing uncertainty in dose
(Urie et a1 1991), increased use of dose-volume histograms, computation of
TCP and NTCP, correlated imaging applied to treatment planning and monitoring,
and real-time visualization of beams and anatomy via the BEV. Despite years of
development there is still a feeling that more effort is needed on implementing
practical tissue-inhomogeneity corrections, especially to account for perturbations
in electron transport. Not surprisingly, non-coplanar field treatments (called
‘unconstrained’ in the NCI contract) have received higher scores than coplanar
field treatments. Perhaps more surprisingly, the use of very high-energy photons
was de-emphasized (Laughlin et a1 1991).
In the next chapter aspects of optimization of conformal radiotherapy will
be discussed, starting with some general two-dimensional considerations and
working towards three-dimensional conformal radiotherapy.
REFERENCES
Aird E G A 1988 Basic physics for medical imaging (Oxford Heinemann)

Arun K S , Huang T S and Blostein S D 1987 Least squares fitting of two 3D
point sets IEE Trans. on Pattern analysis and machine intelligence PAM1-9
(5) 698-700
Bajcsy R and Kovacic S 1989 Multiresolution elastic matching Comp. Vision.
Graphics and Image Processing 46 1-21
Bajcsy R, Reivich M and Kovacic S 1988 Evaluation of registration of PET
images with CT images Technical report MS-CIS-86-89 GRASP LAB 163,
Department of Computer and Information Science, University of Pennsylvania
Bidaut L 1991 Composite PET and MRI for accurate localisation and metabolic
modeling: a very useful tool for research and clinic Proc. SPIE 1445 (Image
Processing) 66-77
Bleehan N M, Glatstein E and Haybittle J L 1983 Radiation therapy planning
(New York: Marcel Dekker)
Boesecke R, Bauer B, Schlegel W, Schad L and Lorenz W J 1987 Medical
workstation for radiation therapy planning The use of computers in radiation
therapy ed I A D Bruinvis et a1 (Proc. 9th ICCRT) pp 181-183
Boesecke R, Doll J, Bauer B, Schlegel W, Pastyr 0 and Lorenz W J 1988
Treatment planning for conformation therapy using a multi-leaf collimator
Strahlentherapie und Oncologie 164 151-154

Brahme A 1992 Recent developments in radiation therapy planning Proc. ART91

(Munich)(abstract book p 6 ) Advanced Radiation Therapy: Tumour Response
Monitoring and Treatment Planning ed A Breit (Berlin: Springer) pp 379-387
Brahme A, Lind B K and Kallman P 1992 Application of radiation-biological
data for dose optimisation in radiation therapy Proc. ART91 (Munich}(abstract
book p 73) Advanced Radiation Therapy: Tumour Response Monitoring and
Treatment Planning ed A Breit (Berlin: Springer) pp 407-415
Broit C 1981 Optimal registration of deformed images Doctoral dissertation
University of Pennsylvania
Burman C, Kutcher G J, Emami B and Goitein M 1991 Fitting of normal tissue
tolerance data to an analytic function Int. J . Rad. Oncol.Biol. Phys. 21 123-135
Chen C T, Ouyang X, Ordonez C, Hu X, Wong W H and Metz C E 1991
Incorporation of structural CT and M R images in PET image reconstruction
Proc. SPIE 1445 (Image Processing) 222-225
Chen G T Y, Kessler M and Pitluck S 1985 Structure transfer between sets of
three-dimensional medical imaging data Proc. Comp. Graph. (Dallas)p 171-
177
Chen G T Y, Pelizzari C A and Levin D N 1990 Image correlation in oncology
Important advances in oncology 1990 ed V T DeVita Jr, S Hellman and S A
Rosenberg (Philadelphia: J B Lippincott CO)pp 131-141
Chen G T Y, Spelbring D R, Pelizzari C A, Balter J M, Myrianthopoulos L C,
Vijayakumar S and Halpern H 1992 The use of beams-eye-view volumetrics in
the selection of non-coplanar radiation portals Int. J . Rad. Oncol. Biol. Phys.
23 153-163
Christiansen H and Stephenson M 1985 MOVIE.BYU (Provo, Utah: Community
Press)
Cunningham J R 1982 Tissue inhomogeneity corrections in photon beam
treatment planning Progress in medical radiation physics I ed C G Orton (New
York: Plenum) pp 103-1 3 1
- 1983 Physics and Engineering perspective Computed tomography in
radiation therapy ed C C Ling, C C Rogers and R J Morton (New York: Raven)
pp 259-26 1
-1988 Computer applications in radiotherapy Selected topics in Physics of
radiotherapy and imaging ed U Madhvanath, K S Parthasarathy and T V
Venkateswaran (New Delhi: McGraw-Hill) pp 3 1-47
Dann R, Holford J, Kovacic S, Reivich M and Bajcsy R 1989 Evaluation of elastic
matching system for anatomic (CT,MR) and functional (PET) cerebral images
J . Comp. Assist. Tomogr. 13 (4) 603-61 1
Dobbs H J 1992 Dose specification: is there a concensus? Proc. 50th Annual
Congress of the British Institute of Radiology (Birmingham, 1992) (London:
BIR) p 4
Drzymala R E, Mohan R, Brewster L, Chu J, Goitein M, Harms W and Urie M
1991 Dose-volume histograms Znt. J . Rad. Oncol. Biol. Phys. 21 71-78
Emami B, Lyman J, Brown A, Coia L, Goitein M, Munzenrider J E, Shank B, Solin

References 57
L J and Wesson M 1991 Tolerance of normal tissue to therapeutic irradiation

Int. J . Rad. Oncol. Biol. Phys. 21 109-122
Ende G, Treuer H and Boesecke R 1992 Optimisation and evaluation of landmark-
based image correlation Phys. Med. Biol. 37 261-271
Evans A C, Dai W, Collins L, Neelin P and Marrett S 1991 Warping of
a computerized 3D atlas to match brain image volumes for quantitative
neuroanatomical and functional analysis Proc. SPZE 1445 (Image Processing)
236-246
Evans A C, Marrett S , Collins L and Peters T M 1989 Anatomical-functional
correlative analysis of the human brain using three-dimensional imaging
systems Proc. SPIE 1092 (Medical Imaging 3: Image Processing) 264-274
Evans P M, Gildersleve J Q, Morton E J, Swindell W, Coles R, Ferraro M,
Rawlings C, Xiao 2 R and Dyer J 1992a Image comparison techniques for
use with megavoltage imaging systems Brit. J. Radiol. 65 701-709
Evans P M, Gildersleve J Q, Rawlings C and Swindell W 1992b The
implementation of patient position correction using a megavoltage imaging
device on a linear accelerator Brit. J . Radiol. in press
Evans P M, Swindell W, Morton E J, Gildersleve J and Lewis D G 1990 A
megavoltage imaging system for patient position reproducibility studies The
use of computers in radiation therapy: Proceedings of the 10th Conference ed
S Hukku and P S Iyer (Lucknow) pp 99-102
Fraass B A 1986 Practical implications of three-dimensional radiation therapy
treatment planning A categorical course in radiation therapy treatment
planning ed B R Paliwal and M L Griem (Oak Brook, IL: Radiological Society
of North America) pp 13-21
-1990 Clinical application of 3D treatment planning AAPM Summer School
1990
Fraass B A and McShan D L 1987 3D treatment planning: 1 overview of a clinical
planning system The use of computers in radiation therapy ed I A D Bruinvis
et a1 (Proc. 9th ICCRT) pp 273-276
Fraass B A, McShan D L, Ten Haken R K and Lichter A S 1990 A practical
3D treatment planning system: overview and clinical use for extemal beam
planning Int. J . Rad. Oncol. Biol. Phys. in press
Fraass B A, McShan D L and Weeks K J 1987 3D treatment planning: 3.
Complete beams-eye-view planning capabilities The use of computers in
radiation therapy ed I A D Bruinvis et a f (Proc. 9th ICCRT) pp 193-196
Fuchs H, Levoy M and Pizer S M 1990 Interactive visualisation of 3D medical
data Visualisation in scientific computing ed G M Nielson, B Shriver and L J
Rosenblum (Los Alamitos, California: IEEE Computer Society Press) pp 140-
146
Gamboa-Aldeco A, Fellingham L L and Chen G T Y 1987 Correlation of 3D
surfaces from multiple modalities in medical imaging SPZE Medical Image
Processing and Display and Picture Archiving and Communications Systems
for Medical Applications pp 1897-1908

Gee J C, Reivich M, Bilaniuk L, Hackney D, Zimmermann R, Kovacic S and

Bajcsy R 1991 Evaluation of multiresolution elastic matching using MRI data
Proc. SPIE 1445 (Image Processing) 226-234
Geiss R A and McCullough E C 1977 The use of CT scanners in megavoltage
photon beam therapy planning Radiology 124 133-141
Gellert W, Kustner H, Hellwich M and Kastner H 1977 The VNR Concise
encyclopedia of mathematics (New York: Van Nostrand) p 708
Gifford D 1984 Handbook of physics for radiologists and radiographers
(Chichester: Wiley)
Goitein M 1982 Applications of computed tomography in radiotherapy treatment
planning Progress in medical radiation physics I ed C G Orton (New York:
Plenum) pp 195-293
- 1983 Patient position during CT scanning Computed tomography in
pp 147-153
- 1985 Calculation of the uncertainty in the dose delivered during radiation
therapy Med. Phys. 12 608-612
-1990 Oblique sections need 3D names also Int. J . Rad. Oncol. Biol. Phys.
19 821
- 1991 The practical incorporation of biological models into treatment
planning Proc. ART91 (Munich) (abstract book p 72)
Goitein M, Abrams M, Rowel1 D, Pollari H and Wiles J 1983a Multidimensional
treatment planning. 2: Beams eye view, back projection and projection through
CT sections Int. J . Rad. Oncol. Biol. Phys. 9 789-797
-1983b A multidimensional treatment planning system Computed tomogra-
phy in radiation therapy ed C C Ling, C C Rogers and R J Morton (New York:
Raven) pp 175-176
Gildersleve J 1991 private communication
Guzzardi R 1987 Physics and engineering of medical imaging (Dordrecht:
Martinus Nijhoff (in collaboration with NATO Scientific Affairs Division))
Hahn P, Shalev S, Viggars D and Therrien P 1990 Treatment planning for protocol-
based radiation therapy Int. J . Rad. Oncol. Biol. Phys. 18 937-939
Hamilton B (ed) 1982 Medical diagnostic imaging systems: technology and
applications (New York: Frost and Sullivan)
Hawkes D J, Hill D L G, Lehmann ED, Robinson G P, Maisey M N and Colchester
A C F 1990 Preliminary work on the interpretation of SPECT images with the
aid of registered MR images and an M R derived 3D neuro-anatomical atlas
NATO ASI Series Vol F 60 (30 imaging in medicine) ed K H Hohne et a1
(Heidelberg: Springer) pp 241-252
Hill D L G, Hawkes D J, Crossman J E, Gleeson M J, Cox T C S , Bracey E E
C M L, Strong A J and Graves P 1991a Registration of MR and CT images
for skull base surgery using point-like anatomical features Brit. J . Radiol. 64
1030-1035
Hill D L G, Hawkes D J and Hardingham C R 1991b The use of anatomical

References 59
knowledge to register 3D blood vessel data derived from DSA with MR images
Pmc. SPIE 1445 (Image Processing) 248-357
Hill D L G, Hawkes D J, Lehmann E D, Crossman J E, Robinson L, Bird C F and
Maisey M N 1990 Registered high resolution images in the interpretation of
radionuclide scans Proc. Annual Conf. of the IEEE Engineering in Medicine
and Biology Society 12 (1) 143-144
Holupka E J, Makrigiorgos G M and Kooy H M 1991 Verification of an image
correlation algorithm using a multi-modality phantom Med. Phys. 18 606
Holupka E J and Kooy H M 1991 A geometric algorithm for multi-modality image
correlation Med. Phys. 18 601
Hopewell J W, Moms A D, Dixon-Brown A 1987 The influence of field size on
the late tolerance of the rat spinal chord to single doses of x-rays Brit. J . Radiol.
60 1099-1108
Horwich A 1990 The future of radiotherapy Radiother. Oncol. 19 353-356
ICRU 1993 Prescribing, recording and reporting photon beam therapy ZCRU
report 50 (Bethesda, MD: International Commission on Radiation Units and
Measurement) in press
Ishigaki T, Itoh Y, Horikawa Y, Kobayashi H, Obata Y and Sakuma S 1990
Computer-assisted conformation radiotherapy Medical Physics Bulletin of the
Association of Medical Physicists ofIndia 15 (3,4) 185-1 89
Jacky J and White C P Testing a 3D radiation therapy planning program Znt. J .
Rad. Oncol. Biol. Phys. 18 253-261
Kallman P, Agren A and Brahme A 1992b Tumour and normal tissue responses
to fractionated non-uniform dose delivery (submitted to Int. J. Rad. Oncol.
Biol. Phys.) and also in Kallman PhD thesis Optimization of radiation
therapy planning using physical and biological objective functions (Stockholm
University, Department of Radiation Physics)
Kallman P, Lind B K and Brahme A 1992a An algorithm for maximising the
probability of complication-free tumour control in radiation therapy Phys. Med.
B i d . 37 87 1-890
Kessler M L 1987 Computer techniques for correlating NMR and X Ray CT
imaging for radiotherapy treatment planning The use of computers in radiation
therapy ed I A D Bruinvis et a1 (Proc.9th ICCRT) pp 4 4 1 4 4
Kessler M L 1989 Integration of multimodality imaging data for radiotherapy
treatment planning PhD thesis University of Califomia at Berkeley
Kessler M L, ten Haken R, Fraass B and McShan D 1991a Expanding the use
and effectiveness of dose-volume-histograms for 3D treatment planning Med.
Phys. 18 611
Kessler M L, Pitluck S , Petti P and Castro J R 1991b Integration of multimodality
imaging data for radiotherapy treatment planning Int. J . Rad. Oncol. Biol. Phys.
211653-1667
Kijewski P 1987 Data dependencies in a three-dimensional treatment planning
system The use of computers in radiation therapy ed I A D Bruinvis et a1 (Proc.
9th ICCRT) pp 53-56

Kom G A and Kom T M 1968Mathematical handbook (New York: McGraw-Hill)

Kutcher G J and Burman C 1989 Calculation of complication probability factors
for non-uniform normal tissue irradiation: the effective volume method Int. J .
Kutcher G J, Burman C, Brewster L, Goitein M and Mohan R 1991
Histogram reduction method for calculating complication probabilities for
three-dimensional treatment planning evaluations Int. J . Rad. Oncol. Biol.
Phys. 21 137-146
Kutcher G J, Burman C and Lyman J 1990 Response to letter by Dr Yaes Int. J .
Laughlin J S , Goitein M, Purdy J A and Sontag M 1991 (Writing Chairs)
Evaluation of high energy photon external beam treatment planning Int. J . Rad.
Oncol. Biol. Phys. 21 3-8
Levin D 1990 MR and PET data merge in 3D images of brain Diagnostic Imaging
(Jan/Feb) pp 28-33
D N Levin, X Hu, K K Tan, S Galhotra, C A Pelizzan, G T Y Chen, R N Beck,
C T Chen, M D Cooper, J F Mullan, J Hekmatpanah and J P Spire 1989 The
brain: integrated three-dimensional display of MR and PET images Radiology
172783-789
D N Levin, X Hu, K K Tan, S Galhotra, A Hermann, C A Pelizzari, G T Y Chen, R
N Beck, C T Chen and M D Cooper 1991 Integrated three-dimensional display
of MR, CT and PET images of the brain SD imaging in medicine ed J K Udupa
and G T Herman (Boca Raton, FL: Chemical Rubber Company) pp 271-283
D N Levin, C A Pelizzari, G T Y Chen, C T Chen and M D Cooper 1988
Retrospective geometric correlation of MR, CT and PET images Radiology
169 8 17-823
Lichter A S 1986 Clinical practice of modem radiation therapy treatment planning
A categorical course in Radiation therapy treatment planning ed B R Paliwal
and M L Griem (Oak Brook, IL: Radiological Society of North America) pp
7-1 2
-1991 Three-dimensional conformal radiation therapy: a testable hypothesis
Lichter A S , Fraass B A, van de Geijn J, Fredrickson H A and Glatstein E
1983 An overview of clinical requirements and clinical utility of computed
tomography based radiotherapy treatment planning Computed tomography in
pp 1-21
Lyman J T 1985 Complication probability as assessed from dose volume
histograms Rad. Res. 104 S-134-19
Lyman J T and Wolbarst A B 1987 Optimisation of radiation therapy 3. A method
of assessing complication probabilities from dose-volume histograms Int. J .
- 1989 Optimisation of radiation therapy 4. A dose-volume histogram
reduction algorithm Int. J . Rad. Oncol. Biol. Phys. 17 433-436

References 61
Margulis A R 1992 Diagnostic imaging in oncology-present and future Proc.

ART91 (Munich) (abstract book p 4,5) Advanced Radiation Therapy: Tumour
Response Monitoring and Treatment Planning ed A Breit (Berlin: Springer) pp
3-7
M m D and Hildreth E C 1980 Theory of edge detection Proc. R. Soc. (London)
207 187-217
McCullough E C 1983 Specifying a CT scanner for use in radiation therapy
planning Computed tomography in radiation therapy ed C C Ling, C C Rogers
and R J Morton (New York: Raven) pp 143-146
McShan D L 1986 Treatment plan evaluation and optimisation A categorical
course in Radiation therapy treatment planning ed B R Paliwal and M L Griem
(Oak Brook, IL: Radiological Society of North America) pp 33-39
-1990a Conformal treatment planning at the University of Michigan The use
of computers in radiation therapy: Proceedings of the 10th Conference ed S
Hukku and P S Iyer (Lucknow) p 80
- 1990b Conformal treatment planning Medical Physics Bulletin of the
Association of Medical Physicists of India 15 (3,4) 190-199
McShan D L and Fraass B A 1987 3D treatment planning: 2. Integration of gray
scale images and solid surface graphics The use of computers in radiation
therapy ed I A D Bruinvis et a1 (Proc. 9th ICCRT) pp 41-44
-1990 A practical 3D treatment planning system: anatomy and display Znt. J .
Rad. Oncol. B i d . Phys. in press
McShan D L, Fraass B A and Lichter A S 1990 Full integration of the beam’s eye
view concept into computerized treatment planning Znt. J . Rad. Oncol. Biol.
Phys. 18 1485-1494
McShan D L, Matrone G M, Fraass B A and Lichter A S 1987 A large screen
digitiser system for radiation therapy planning Med. Phys. 14 (3) 459
McShan D L, Silverman A, Lanza D M, Reinstein L E and Glicksman A S 1979
A computerized three-dimensional treatment planning system utilizing colour
graphics Brit. J . Radiol. 52 478-48 1
Moerland M A, Bhagwandien R, Beerama R and Bakker C J G 1991
Determination of geometric distortions in M R imaging; implications for
radiotherapy treatment planning Proc. 1st biennial ESTRO meeting on physics
in clinical radiotherapy (Budapest 1991) p 25
Moores B M, Parker R P and Pullan B R (eds) 1981 Physical aspects of medical
imaging (Chichester: Wiley)
Nahum A E and Tait 1992 Maximising local control by customized dose
prescription for pelvic tumours Proc. ART9I (Munich) (abstract book p 84)
Advanced Radiation Therapy: Tumour Response Monitoring and Treatment
Planning ed A Breit (Berlin: Springer) pp 425431
Neiw H M, Chen C T, Lin W C and Pelizzari C A 1991 Automated
three-dimensional registration of medical images Proc. SPIE 1445 (Image
Processing) 259-264
Niemierko A and Goitein M 1991 Calculation of normal tissue complication

probability and dose-volume histogram reduction schemes for tissues with a

critical element architecture Radiother. Oncol. 20 166-176
Orton C G 1988 A unified approach to dose-effect relationships in fractionated
radiotherapy Selected topics in Physics of radiotherapy and imaging ed U
Madhvanath, K S Parthasarathy and T V Venkateswaran (New Delhi: McGraw-
Hill) pp 25-30
Parker R P and Hobday P A 1981 CT scanning in radiotherapy treatment planning:
its strengths and weaknesses Computerised axial tomography in oncology ed J
E Husband and P A Hobday (Edinburgh: Churchill Livingstone) pp 90-100
Parker R P, Hobday P A and Cassell K J 1979 The direct use of CT numbers in
radiotherapy dosage calculations for inhomogeneous media Phys. Med. Biol.
24 802-809
C A Pelizzari and G T Y Chen 1987 Registration of multiple diagnostic imaging
scans using surface fitting The use of computers in radiation therapy ed I A D
Bruinvis et a1 (Proc. 9th ICCRT) pp 4 3 7 4 0
C A Pelizzari, G T Y Chen, D R Spelbring, R R Weichselbaum and C T Chen
1989 Accurate three-dimensional registration of CT, PET, and/or MR images
of the brain J . Comp. Assist. Tomog. 13 20-26
Phillips M H, Kessler M, Chuang F Y S , Frankel K A, Lyman J T, Fabrikant J I
and Levy R P 1991 Image correlation of MRI and CT in treatment planning for
radiosurgery of intracranial vascular malformations Znt. J . Rad. Oncol. Biol.
Phys. 20 881-889
Press W H, Flannery B P, Teukolsky S A and Vatterling W T 1988 Numerical
recipes: The Art of Scientific Computing (Cambridge: Cambridge Univ Press)
p 52 et seq
Purdy J A, Wong J W, Harms W B, Drzymala R E, Emami B, Matthews J W,
Krippner K and Ramchandar P K 1987 Three dimensional radiation treatment
planning system The use of computers in radiation therapy ed I A D Bminvis
et al (Proc. 9th ICCRT) pp 277-279
Reichenberger H 1990 Biomagnetic diagnostics Imaging systems for medical
diagnostics ed E Krestel (Berlin and Munich: Siemens AG)
Reinstein L E, McShan D L, Land R E and Glicksman A S 1983 Three-
dimensional reconstruction of CT images for treatment planning in carcinoma
of the lung Computed tomography in radiation therapy ed C C Ling, C C
Rogers and R J Morton (New York: Raven) pp 155-165
Reinstein L R, McShan D L, Webber B M and Glicksman A S 1978 A computer-
assisted three-dimensional treatment planning system Radiology 127 259-264
Rizzo G, Gilardi M C, Bettinardi V, Carutti S and Fazio F 1990 Integration of
multimodal medical images Proc. Annual Conf. of the ZEEE Engineering in
Medicine andBiology Society 12 (1) 145-146
Roeske J C, Pelizzari C A, Spelbring D, Blend M, Belcaster G and Chen G
T Y 1991 Registration of SPECT and CT images for radiolabelled antibody
biodistribution analysis Med. P hys. 18 649
Rosenman J 1991 3D imaging in radiotherapy treatment planning 3D imaging in

References 63
medicine ed J K Udupa and G T Herman (Boca Raton, FL: Chemical Rubber

Company) pp 3 13-329
Sailer S L, Bourland J D, Rosenman J G, Sherouse G W, Chaney E L and Tepper
J E 1990 3D beams need 3D names Int. J . Rad. Oncol. Biol. Phys. 19 797-798
Salehpour M R, Pelizzari C A , Balter J M and Chen G T Y 1991 A multi-structure
approach to image correlation Med. Phys. 18 600
Schlegel W 1987 Progress and trends in computer assisted radiotherapy Computer
Assisted Radiology ed H U Lemke, M L Rhodes, C C Jaffee and R Felix (Berlin:
Springer) pp 295-299
Schlegel W, Boesecke R, Bauer B, Alandt K and Lorenz W J 1987 Dynamic
therapy planning The use of computers in radiation therapy ed I A D Bruinvis
Schonemann P H 1966 A generalized solution of the orthogonal Procrustes
problem Psychometrika 31 (1) 1-10
Schultheiss T E, Dixon D 0, Peters L J and Thames H D 1987 Calculation of late
complication rates for use in optimisation algorithms The use of computers in
radiation therapy ed I A D Bruinvis et a1 (Proc. 9th ICCRT) pp 127-129
Schultheiss T E, Orton C G and Peck R A 1983 Models in radiotherapy: volume
effects Med. Phys. 10 410-415
Shalev S, Hahn P, Bartel L, Themen P and Carey M 1987 The quantitative
evaluation of alternative treatment plans The use of computers in radiation
therapy ed I A D Bruinvis et a1 (Proc.9th ICCRT) pp 115-118
Shalev S , Viggars D, Carey M and Hahn P 1991a The objective evaluation of
alternative treatment plans. 2: score functions Znt. J . Rad. Oncol. Biol. Phys. 20
1067-1073
Shalev S , Viggars D, McCracken D and Gatschuff J 1991b The limitations
imposed on conformal treatment planning by patient movement and set up
errors Proc. ART91 (Munich) (abstract book p 67)
Sherouse G W, Bourland J D, Reynolds K, McMurry H L, Mitchell T P and
Chaney E L 1990 Virtual simulation in the clinical setting: some practical
considerations Int. J . Rad. Oncol. Biol. Phys. 19 1059-1065
Sherouse G W and Chaney E L 1991 The portable virtual simulator Int. J . Rad.
Smith A R 1990 Evaluation of new radiation oncology technology Int. J . Rad.
Sontag M R, Altschuler M D, Bloch P, Reynolds R A, Wallace R E and Waxler
G K 1987 Design and clinical implementation of a second generation three-
dimensional treatment planning system The use of computers in radiation
therapy ed I A D Bruinvis et a1 (Proc.9th ICCRT) pp 285-288
Sontag M and Purdy J A 1991 (Writing Chairs) State-of-the-art of extemal photon
beam radiation treatment planning Int. J . Rad. Oncol. Biol. Phys. 21 9-23
Swindell W and Gildersleve J Q 1991 Megavoltage imaging in radiotherapy Rad.
Magazine 17 (196) 18-20
Suit H D and Verhey L J 1988 Precision in megavoltage radiotherapy Megavoltage

Radiotherapy 1937-1987 (Brit. J. Radiol. Supplement 22 pp 17-24)

Suit H D and Westgate S J 1986 Impact of improved local control on survival Int.
J . Rad. Oncol. Biol. Phys. 12 453458
Tait D M and Nahum A E 1990 Conformal therapy Eur. J . Cancer 26 750-753
Ten Haken R K, Kessler M L, Stem R L, Ellis J H and Niklason L T 1991 Quality
assurance of CT and MRI for radiation therapy treatment planning Quality
assurance in radiotherapy physics ed G Starkschall and J Horton (Madison,
WI: Medical Physics Publishing) pp 73-103
Tepper J E and Shank B 1991 (Writing Chairs) Three-dimensional display in
planning radiation therapy: a clinical perspective Int. J . Rad. Oncol. Biol.Phys.
21 79-89
Urie M M, Goitein M, Doppke K, Kutcher K G, LoSasso T, Mohan R,
Munzenrider J E, Sontag M and Wong J W 1991 The role of uncertainty
analysis in treatment planning Int. J . Rad. Oncol. B i d . Phys. 21 91-107
Viggars D A and Shalev S 1990a Dose optimisation in 3D treatment planning The
S Hukku and P S Iyer (Lucknow) pp 308-31 1
-1990b Colour graphics for 3D treatment planning The use of computers in
radiation therapy: Proceedings of the 10th Conference ed S Hukku and P S
Iyer (Lucknow) pp 1 11-1 14
Wambersie A, Hanks G and Van Dam J 1988 Quality assurance and accuracy
required in radiation therapy: biological and medical considerations Selected
topics in Physics of radiotherapy and imaging ed U Madhvanath, K S
Parthasarathy and T V Venkateswaran (New Delhi: McGraw-Hill) pp 1-24
Webb S (ed) 1988 The physics of medical imaging (Bristol: Adam Hilger)
-1990 Non standard CT scanners: their role in radiotherapy Int. J . Rad. Oncol.
B i d . Phys. 19 1589-1607
-199 1 Optimisation by simulated annealing of three-dimensional, conformal
treatment planning for radiation fields defined by a multileaf collimator Phys.
Med. Biol. 36 1201-1226
Wells P N T (ed) 1982 Scientific basis of medical imaging (Edinburgh: Churchill
Livingstone)
Wheldon T 1988 Mathematical models in cancer research (Bristol: Adam Hilger)
Wong J W 1991 Imaging in radiotherapy Proc. 1st biennial ESTRO meeting on
physics in clinical radiotherapy (Budapest, 1991) p 17
Wong J W and Purdy J A 1990 Advances in radiotherapy treatment planning and
verification The use of computers in radiation therapy: Proceedings of the 10th
Conference ed S Hukku and P S Iyer (Lucknow) pp 72-75
Yaes R J 1990 Volume effects Int. J . Rad. Oncol. Biol. Phys. 18 975-6
Yu X and Chen G T Y 1991 Applying correlated CT and MR images of the brain
in compensation for attenuation in PET image reconstruction Med. Phys. 18
649

CHAPTER 2
TREATMENT PLAN OPTIMIZATION
2.1. GENERAL CONSIDERATIONS
There are many aspects of delivering external beam radiotherapy which should
be addressed when attempting treatment-plan optimization. For the moment,
however, and for the purposes of introducing the problem, let us suppose that a
target volume has been identified by an imaging method and consider a simple
transaxial slice of the patient in which this target volume reduces to a simple
target area, possibly with some simple, for example circular, outline. The ideal
aim in this reduced version of the 3D planning problem would be to deliver a
specified uniform dose to that area and to deliver zero dose elsewhere, clearly
impossible since to reach the target area, photons must travel across surrounding
tissue depositing dose en route. Hence the aim is generally modified to that of
delivering a specified high dose to the target area and as low a dose as possible
elsewhere. Let us look at how this could be done by rotation therapy i. Instead of
the conventional small number of fixed flat or wedged fields, imagine that the beam
is deliverable from a full range of orientations in G2n. Furthermore imagine that
the profile of the beam may be varied, i.e. the intensity may be non-uniform over
the width of the beam port.
Set up a coordinate system as shown in figure 2.1, called dose space, with dose
points specified by d (r, 8). Define the origin of this space as also the axis of
rotation of the treatment machine. Let f (r+,4) specify the intensity of the beam
at angle 4 and at a distance r+ from the projection of the axis of rotation, along
the beam. f (r+,4) is non-zero only for postive r#. The normal from the origin
to this projection is the first ray of the beam and the beam is blocked for negative
r4*
t Nowadays, rotation therapy, common many years ago (see appendix 2A), has fallen
rather out of favour. This is possibly because using a small number of fixed fields allows
sensitive organs to be out of the shadow of the primary beam completely, whereas, unless
synchronous protection is allowed (see section 2.4), all tissue receives some dose in rotation
therapy. However, consideration of rotation therapy provides a good insight into the dose-
planning problem and is a natural way to begin to consider solving the inverse problem of
determining the beam intensitiesfrom a dose prescription.
65
66 Treatment plan optimization
Figure 2.1. Showing the dose space and the relationship between dose at a
point (r, 6) and the beam intensity f ( r # , # ) at a particular position r4 in the
treatment beam port at angle 4.
Now imagine for the moment that the beam is unattenuated on passage through
the patient (ie is of sufficiently high energy). Then we observe that the dose
d (r, 8 ) is proportional to the intensity f ( r cos(8 - 4), 4) of the beam at angle
4. Again for simplicity we shall ignore proportionality (and conflicts of units and
dimensions) and say the dose is this intensity. After a full rotation through 2 n the
dose will be
because there will only be contributions to the dose d (r,e) for the half of the 277
rotation when this dose point is in the unblocked beam. Because d (-r, 8 n)= +
d (r, e), we may write more simply
(2.lb)
4
where the extra factor is included because this range evaluates the dose to each
point twice. (We shall return later to the apparent paradox of ignoring attenuation,
in which case there would be no photon collisions in the material at which to
deposit dose.) Equation (2.1b ) is the familiar Radon Transform, the relationship
which, in x-ray computed tomography (CT), expresses a 2D quantity in terms of
its 1D projections (Herman 1980). Once again, as in CT, the unknown quantity is
under the integral sign whereas the known quantity (here taken to be known in the
sense of a specified dose prescription) is to the left of the equality, i.e. given a dose
prescription d (r, 8) we require to compute the distribution of profrles f (r#,4).
We need an inverse Radon Transform. This similarity between reconstruction
tomography and optimized dose planning has been pointed out by a number of
authors including Cormack (1987), Cormack and Cormack (1987), Cormack and
Quinto (1989), Brahme (1988), Webb (l989,1990a),Bortfeldeta1(1990a,1992),

The impossibility of true inverse computed tomography 67
Jones ( 1 990) and Lind and Kallman (1990). It is appropriateto call optimized dose
planning ‘inverse computed tomography’.
The degree of approximation in equation (2.lb) depends on the energy of the
beam. If a very fine beam of monoenergetic radiation falls collinearly on opposite
sides of a slab of material of attenuation coefficient p , and r is the distance from
the centre of the slab, the intensity of primary radiation at r is proportional to
coshpr. For 10 MeV radiation, p N 0.02 cm-’, so for a slab 30 cm thick, the
largest value of p r is 0.3. As cosh0.3 = 1.045, the zero-order approximation
( p 2: 0) is correct at this energy to better than 5%. For 6oCoradiation, however,
the error would be 50%. The zero-order approximation also ignores electronic
disequilibrium near the surface and also scatter.
Cormack (1987)provides an argument that, for parallel-opposed pencil beams,
the isodose curves including scatter will be cylinders concentric with the line of
the beams, in which case equation (2.l b ) holds provided f inside the integral is
replaced by a function fs where
and g (s) is the dose at a distance s from the primary beam line.
The p = 0 approximation is a simplification developed to allow a discussion of
the impossibility of true inverse computed tomography, even under this unphysical
assumption. In section 2.3 we shall retum to what can be analytically deduced
when the p # 0 condition is restored.
2.2. THE IMPOSSIBILITY OF TRUE INVERSE COMPUTED

TOMOGRAPHY
Some might feel that it is obvious that equation (2.lb) cannot be inverted for
the most general real dose prescription. Indeed this is true. However, a nice
intuitive demonstration of the difficulties has been provided by Cormack (1987)
and is summarized here. We can draw the ‘f -space’ f (r#,4) in two dimensions
(figure 2.2). This might also be called the ‘beam element space’. Draw a line
from the origin in f -space to a point T a distance r away at an angle 8. Then
construct a circle on this diameter OT and any triangle OTQ where Q is also on
the circumference. It is now clear that the right-hand side of equation (2.lb) is
simply proportional to the mean of the values off around the circumference of this
circle (for a full rotation of the beam implemented via equation (2.l b ) the point Q
actually goes twice round the circle). An ideal arrangement for delivering a high-
dose point would be for f to be unity on this circumference and zero elsewhere.
Now consider (figure 2.3) trying to deliver a uniform dose to a small circular
area in dose space, a distance h from the origin and of radius a. If we place this
area in beam space (figure 2.4), then we can construct circles on all the diameters

68 Treatmentplan optimization
1 beam space ‘f’
Figure 2.2. Showing f -space or beam-element space. Integration of f

around the circumference of the circle produces the dose at (r,0). This
corresponds to rotation of the beam shown in figure 2.1 around a half circle.
7 d - space
Figure 2.3. A target region of high dose in dose space. The region is shown
as a circular area of radius ‘a’ and centre a distance ‘h’from the origin.
Figure 2.4. The corresponding area in f-space contributing to the

high-dose region shown infrgure 2.3 (shown hatched). The area is bounded
by a limaEon of Pascal.
having ends at the origin, within, and on, the circumference of this circular area.
It follows that the beam intensity requires to be non-zero around all these circles
and zero elsewhere. The envelope of these circles is the limaqon of Pascal
in acoordinate system where the centre of the target region of radius a is at distance
h from the origin along the x-axis. In principle, too, this is no problem.
Aside: the envelope is constructed as follows (refer tofigure 2.5). Let C be any
point on the circumference of the circle. The envelope is then the pedal curve, with
respect to the origin, swept out by the intersection of the normal to the tangent at

f-space
~ ‘ j
Figure 2.5. Showing the construction of the envelope of the circles passing
through the circular dose region (see text for symbolsfor figures 2.5 to 2.7).
C , i.e. the locus of the point which is the intersection of the tangent to C and the
perpendicular to this tangent which passes through the origin. In figure 2.5(a)
the pedal is on the inner loop of the limacon and in figure 2 S ( b ) it is on the outer
loop. The pedal is at the origin when the tangent to C also passes through the
origin; this is the transition from inner to outer loop. This gives the polar equation
r@= Ih cos 8 - a I (Cormack 1987)t.
Now consider, as happens in practice, that there is another region (an ‘organ at
t It may not be immediately obvious why the pedal curve is the envelope. Kom and Kom
(1968)show that the envelope of any one-parameter family of plane curves described by
$ (x, y, A) = 0 is obtained by eliminating the parameter h between this equation and its
partial derivative a$ (x, y , A) /ah = 0. Consider figure 2.6 in which (rb,a ) is a general
point on the circumference of the circle whose diameter is OC and the line from C to the
origin of the circular area of high dose makes an angle B with the line joining the origin to
the centre of that circular area. The equation of the circle is then
rb + a sina sinB - cosa ( h - a cos j3) = 0 (2.3)
The ‘one parameter’ to be eliminated is then B. Differentiating this equation and setting
the result to zero gives
a s i n a c o s @- a c o s a s i n @ = O (2.4)
from which (Y = B, i.e. the line joining C to the point (r,+,,
a ) is a tangent. Putting (Y =3!,
in the general equation (2.3) gives the limaGon r,+,= h cosa - a.

Figure 2.6. Showing the construction for the IimaGon.
Figure 2.7. High- and low-dose regions in dose space.
,, ,>
f-space
Figure 2.8. The corresponding area in f -space contributing to the high-

and low-dose regions shown in figure 2.7. The hatched region is the area of
overlap.
risk’) which it is required to shield, geometrically separated from the first region
in space. Suppose for argument this is also a circle (figure 2.7). Now imagine
placing this circle also in beam-space and making the same construction as before
(figure 2.8). The difficulty is that the two limaqons intersect. A conflict arises. The
region inside the limason embracing the high-dose target region requires non-zero
intensity f , whereas the region inside the limaqon embracing the sensitive region
(organ at risk) requires zero f . This is clearly impossible.
Thus Cormack argued that in general the inverse tomography problem is
insoluble and confirms intuition. There is in general no distribution of beam
intensities which can give a prescribed dose distribution exactly, even when the

assumptions are as simple as those going to make up equation (2.1b). Cormack

calls this the zero-order approximation: it does not reflect reality but it is a good
start allowing us to see that even then the problem is insoluble. Sometimes
it is stated that the problem would be soluble if there could be negative beam
intensities, but this is obviously unphysical. Cormack (1987) gives a solution
to inverting equation (2.1b ) involving negative beam weights. He then suggests
simply adding a constant so as to make the smallest beam weight zero. This has a
certain elegance, but in a sense the real dose-planningproblem has not been solved.
The treatment-planning physicist requires a solution with non-zero beams.
Thus we appreciate that in speaking of ‘solving’ the inverse problem we mean
obtaining an approximation to the solution. Often this is done by iterative methods
(see section 2.5.4).
If the patient is considered to have a circular outline of radius R and one
cannot make the assumption that cosh p r is 1 , then the following equation (2.5)
representing a first-order approximation replaces equation (2.Ib). The dose space
is considered to be irradiated with parallel-opposedfields on one side of the normal
through the origin (other side blocked to irradiation), each with the same (but
times half the magnitude) profile of intensity and a full 21r rotation (Cormack and
Cormack 1987), i.e. the irradiation is as in figure 2.1 with non-zero f (r#,4) for
positive r+ only:
x exp ( - p ( R 2 - r:) ’) cosh ( p r sin (8 - 4)) d4
where r# = r cos (e - 4). Notice the range of integration, which at first sight
might appear to be only a half-rotation of the parallel-opposed beams. This range
does represent a full 2n rotation. For any particular point d (r, 6 ) there is a
contribution only from those profiles at orientations 8 - (5) < < +
4 8 (5)
because for other onentations the dose point is in the line of sight of the block.
Putting p = 0 reduces equation (2.5) to equation ( 2 . 1 ~ ) The . same argument
can be made regarding scatter. The problem is now to solve the attenuated Radon
Transform problem. This cannot in general be done unless very restricting further
(and unphysical) approximations are made (Brahme et a1 1982, Lax and Brahme
1982, Cormack and Cormack 1987). This author’s view is that, elegant though the
mathematics become, it soon loses touch with reality and if it is desired to find a
solution for a real physical problem, it is necessary to resort to numerical methods
and in particular iterative solutions. These also of course do not solve the problem,
but they deliver an approximate solution for the beam weights corresponding to a
real dose distribution and they can constrain these beamweights to be positive.

2.3. THE CASE OF A CIRCULARLY-SYMMETRIC DOSE DISTRIBUTION
Now assume the beam profiles at each orientation of the beam are all the same (but
still spatially varying along their length), giving as a result a circularly-symmetric
dose distribution. Imagine the situation (figure 2.9(a)) where a complete rotation
of the single shaped beam shown through 277 takes place. The isocentre is the
centre of the circular object. As in all discussions so far, the beam is non-zero
for only positive r4 and so must complete a full rotation to generate a circularly-
symmetric dose distribution. The dose at radius r is given by:
d (r)= (t) /around circle radius r

f ( x = r cos@)
(2.6)
x exp(-p ( R 2 -xZ)')exp(-pz)d@
where the x-axis is defined so that x = r cos4 and the scattered radiation is
ignored. Equation (2.6) does not result from equation (2.5) because the latter was
for a rotation with parallel-opposed fields whereas equation (2.6) is for a rotation
with a single field (Brahme et a1 1982). Now noting that
dr = -r sin @@ (2.7a)
i.e.
d@= -dr ( r 2 - x 2 -10
) (2.7b)
and writing the beam profile f ( x ) in terms of a corresponding dose profile m ( x )
through the origin of dose space, i.e.
and also breaking up the circular integral into partst, and changing to an integral
over x , we obtain
t The simplest way to obtain equation (2.9) is to note that the dose d ( r ) from a full 27c
rotation of the shaped beam is exactly the same as keeping the beam fixed and integrating
the dose profile around the circle on radius r . From figure 2.9(n) it is then clear that
the integration only needs to be performed over the upper semicircle because the beam
< <
is blocked for negative x . The integral is split into two parts for -n/2 q4 0 (negative
< <
z ) and for 0 9 n/2 (positive z ) . When combining these two parts, and changing from
4 limits to x limits, to obtain equation (2.9) care must be taken with the sign of the root
I
2 = (r2-x2)?

The case of a circularly-symmetric dose distribution 73
Figure 2.9. ( a )The symmetric rotation therapy case showing thefield to be

non-zero one side of the projection of the rotation axis and blocked the other
side (adapted fiom Brahme et a1 (1982)).(b)The experimental arrangement
when it is required to shield the region r < ro, in which case the block extends
across the projection of the rotation axis (from Brahme et a1 (1982))(see text
for discussion).
Equation (2.9) was derived by Brahme et a1 (1982) (their equation 3(-)

refer to figure 2.9(a) adapted from that paper for further clarification. Notice
the interesting way of relating the dose distribution at a radius r in terms of the
distribution m (x) from each projection, along a central axis rather than along the
beam port. This is reminiscent of the way of handling projections in x-ray CT
when projections are often transformed to axes through the origin of rotation in
working out the reconstruction mathematics (e.g. Lakshminarayanan 1975). Of
course, when equation (2.9) is solved for m (x) for some particular d ( r ) ,it is still
necessary to correct back to the beam profile f (x) using equation (2.8) above.
Solving equation (2.9) is of interest despite the limiting assumptions of a
circular patient outline, ignoring build-up and scatter. Brahme posed the question:
what is the solution for the general case where m ( x ) is zero inside a radius ro? i.e.
d ( r ) is quite zero inside the circle of radius ro. The problem is then to invert
i.e. d ( r ) is a given function and m (x) is to be determined.

First make the change of variables
x2 = t + r o r2=y+ri (2.11)
which transforms equation (2.10) into a convolution equation
R (y) = l y e O) f (y - t>dt (2.12)
where
(2.13)

g (Y) = d ( ( Y + r02y2) (2.14)

and
cosh ( p (y - t)’/’)
f (Y - t>= (2.15)
(y - r)l/z
It is now possible to Laplace transform the convolution equation. The Laplace
transform is defined by
H (SI = I” exp (-SY> (Y) dY (2.16)
where an upper-case letter represents the Laplace transform of the corresponding

lower-case parameter. So
(2.17)
Now make the assumption that dose is to be a constant D outside r = ro, then the
Laplace transform of dose is
G (s) = D / s . (2.18)
By direct integration of f (y) from equation (2.15) via equation (2.16)
F (s) = (n/s)’/’exp (p2/4s) (2.19)
and so
E exp (-p2/4s)
(SI = G (SI (s/n)’/’ . (2.20)
Inverting the Laplace transform and putting back the original variables we come
to the final result
D X cos ( p (x2 - ri)1’2)
m (x) = (2.21)
(x2 - ri>*”
Some interesting results immediately follow:
1. Suppose we want to have a uniform dose from the origin to a distance rl and
as small a dose as possible for larger r. Simply put ro = 0 in equation (2.21) to
obtain
m ( x ) = D cos ( k x ) forx rl < (2.22~)
and
m (x) =0 for x > rl (2.22b)
i.e. the lateral dose distribution should decrease quite slowly as the cosine of the
distance. Beyond this the dose profile should be set to zero. The delivered dose is
then given by putting equations (2.22a), (2.22b) into equation (2.9), i.e. the dose
at a radius r , d ( I ) , is given by
r (
m ( x ) 2 cosh p ( r z - x’) ”’) dx
(2.9)
n (r2 - x2j1’2

The case of a circularly-symmetric dose distribution 75
with
m ( x ) = Dcos ( p x ) forx < rl (2.224
and
m (x) =0 forx > rl. (2.22b)
The integral can be numerically computed ford ( r ) . This would show a uniform
dose D up to the radius rl , and for r > rl the dose falls off with increasing radius
due to the fact that radiation has to pass through the outer annulus r > rl to get
to the circular high-dose area r 6 rl. The curve shows that the dose falls to half
the value of the dose in the uniform region at about 1.4rl. For very large (and
unphysical) values of r , like r 2 50 cm, the dose profile begins to rise again
because of the behaviour of the cosh term. This need not concern us.
Now we may imagine making rl very small. Equation (2.9) then defines
the point irradiation distribution for a full 2 n rotation irradiation by a pencil
beam. This is a very important fundamental distribution and can form the basis
of a method of dose planning by techniques of ‘inverse computed tomography’
(Brahme 1988, Lind and Kallman 1990). We retum to this in detail in section 2.5.
2. To shield a circular region inside r = ro, simply use equation (2.21). Figure
2.9(b) shows the field arrangement with the block extending beyond x = 0 to
x = ro. Equation (2.21) shows the linear dose profile theoretically rising to
infinity as x approaches ro; Brahme et a1 (1982) state that in practice this infinity
can be simply ‘large’. The equation (2.21) can be realized in practice by having
a central absorbing pin and symmetrically-placedwedge filters (with their narrow
ends touching the pin) to correspond to the profile in equation (2.21) corrected
back to the beam space by equation (2.8). Recently this has been engineered by
Casebow (1990a) (see section 2.4). If the patient were circular, a half-rotation
with such a wedge-plus-pin filter would suffice; in practice a full rotation would
be needed if the field were half-blocked as in figure 2.9(b).
Thus it is possible to arrange for a zero-dose region to lie inside a region of
uniform dose (primary beams only being considered) in this special rotationally-
symmetric geometry. Cormack’s argument is not violated because this stated that
in general the problem is not always soluble for arbitrary dose regions, one of
which does not contain the other. It can easily be seen that for the two concentric
regions here (zero dose within r = ro and uniform dose within r = rl, but outside
r = ro) the limaGons become the circles themselves in f-space (h = 0 in the
general equation (2.2)).All that is required then is to arrange zero f-values within
the circle r = ro and non-zero f-values between the annulus r = ro and r = rl.
Then any circle, such as that shown in figure 2.2 built on a diameter connecting the
origin to any point on the circumference of the r = rl circle, partially lies in the
inner circular region and partially lies in the outer circular annulus. Interestingly,
for all such circles created for values of r between r = ro and r = r l , the f-values
in the non-zero region, with Brahme’s solution, sum to the same value (because
the dose is uniform in this region).

Brahme et a1 (1982) were the first to state the similarity between this problem
and the problem of reconstruction in x-ray CT. The problems are mirrors of
each other, The mathematics is a nice insight into the problem even though the
assumptions are rather limited (no scatter, circular patient, circular treatment area
(or shielded area)). It gives directly the point irradiation distribution.
When the dose distribution is not constant with r an analytical expression can
still be obtained form ( x ) (Brahme et a1 1982) and is
In section 2.5.1 we describe how Lind and Kallman (1990) used this expression to
compare with the result from experiment and also a method of inverse computed
tomography which is actually not restricted to cylindrically-symmetric dose
distributions.
Retuming to the result in equation (2.22) we saw that the dose will not suddenly
reduce from the uniform value within rl to zero outside because of the passage of
photons through the outer region beyond rl . If we really did want this to happen
the intensity function m ( x ) would have to be negative for r > rl. Recently Barth
(1990) has given the appropriate formalism. For the purposes of the argument to
follow, just for the moment assume that these negative intensities are possible-
i.e. suspend disbelief. In that case Barth (1990) shows how the formalism in
equation (2.21a) can be adjusted so that the dose distribution produced is radially-
symmetric about some point other than the rotational axis of the beam. Essentially
this involves a translation and a term to account for the different path lengths of
the radiation. Then he argues that because everything is linear one could create
say two such circular distributions of uniform dose with zero dose elsewhere. This
simply involves adding two intensity profiles varying with angle. Furthermore it
is then possible to imagine that any concave area, within which uniform dose is
required, can be decomposed as a set of (possibly a very large number of) circular
areas with different centres and different radii. So in principle by superposition
of the computations for each of these, a uniform dose can be constructed in any
arbitrary, and if needed concave, area (figure 2.10). The mathematics are elegant.
All so well and good except for the problem of needing negative beam-weights.
When these are set to zero, the dose distribution ceases to be zero outside the
required area and linearity breaks down. Studies such as that of Barth (1 990) are
very interesting philosophically but cannot really address the real-world problems.
Analytic mathematics has also been limited to scatter-free calculations. It is
these limitations which iterative, and admittedly computationally expensive, dose-
planning algorithms can effectively remove. As we have seen, it is apparent that,
without unphysical negative intensities, perfect dose distributions can never be
achieved to exactly match prescription. It is interest in seeing how close one can
get which fuels the problem solving.

Primitive blocked rotation therapy 77
Figure 2.10. A convex phantom of arbitrary cross section with a concave

dose distribution. The arbitrary concave dose distribution is built up out of
N small radially-symmetric non-intersecting dose distributions of diperent
radii, each centred about a diperent point. By invoking negative beam
intensities zero dose can be obtained outside the concave area (from Barth
( I 990)). (Reprinted with permission from Pergamon Press Ltd, Odord, UK.)
Goitein (1990) provides a useful critique of the problem posed by negative beam
intensities. When the inverse problem is solved to give spatially non-uniform
beams, followed by setting negative intensities to zero, Goitein asks in what sense
can the resulting dose distribution be considered optimum? It is certainly not
the true solution. Indeed the trade-off between dose delivered outside the target
volume and dose uniformity inside the target has not been addressed and may not
be optimal. Methods of solving the inverse problem which constrain the beam
intensities to be positive (e.g. Webb 1989) avoid this difficulty altogether (see
section 2.5.4) by seeking practical solutions which minimize some cost function
based on the difference between dose prescription and dose delivered to both target
volume and organs at risk. Cormack (1990) argues that such methods should at
least be given a chance to prove themselves, given work on the inverse problem
is relatively new in the near-100 year history of radiotherapy dose planning.
2.4. PRIMITIVE BLOCKED ROTATION THERAPY
For many years treatment planning ignored the possibility of calculating

f (re, #)t,but recognized that continuous rotation of an open field about a suitable
isocentre would give high (if non-uniform) dose distributions sparing normal
tissue somewhat. Rotation therapy was improved by a technique of shielding
a particular and localized circular structure by suspending a highly absorbing
cylinder in the radiation field upon a plate which pivoted under gravity. This
t The impetus for such calculations really comes from the ‘post-cr era’ from those who
understood well the mathematics of cr.

Figure 2.11. Synchronous protection in rotation therapy using an absorbing

rod (shown hatched). The shaded region ‘p’ is always in the shadow of the
gravity-supported circular block in the field. The dose to the crescent shaped
region ‘U’will be non-uniform. The rod hangs at a distance ‘t’ below the line
from the source to the isocentre. Thisprojects to a distance ‘s’at the isocentre
(see text for details). (From Casebow (1990a)J
is illustrated in figure 2.11. The shielded area ‘p’ lies below the isocentre by a
distance ‘s’. The area treated is ‘v’. The absorbing pin is ‘a’, pivoting at ‘c’.
The distance ‘t’ of the pin below the pivot is adjusted so the magnified distance
measured from the isocentre becomes ‘s’. The plate ‘b’ can rotate freely under
gravity. The point ‘c’ lies on the axis of the treatment beam.
It can be appreciated easily that this arrangement will shield the circular area
shown hatched. By changing the size and offset of the absorbing pin, the size and
offset of the shielded region changes. The treated volume is crescent shaped. This
idea of gravity-assisted blocking was first suggested by Proimos (1961,1963) and
Proimos et a1 (1966). Takahashi et a1 (196 1) developed similar rotation techniques
to those of Proimos. Possibly these were inspired by Takahashi’s pioneering
work in radiological tomography in the pre-CT era (see Webb 1990~).Two main
problems arise:
1. The shielding is limited by the finite size and the attenuation coefficient
of the pin. To obtain the same shielding (for a relatively smaller region) from a
relatively smaller pin, the material would have to change to one of a higher linear
attenuation coefficient. Proimos tried using gold, lead and platinum absorbers.
2. The rod absorber perturbs the dose uniformity in the exposed region.
To some extent problem 1 can be overcome by using a thicker block and
gearing arrangements whereby, as well as gravity positioning, the block is tumed
synchronously with gantry rotation to present a profile of fixed absorbing depth
whatever the gantry orientation (Proimos 1961), i.e. the lead blocks remain
parallel to the direction of the rays whatever the gantry position. For lead blocks
with thickness in excess of some 7 cm the direct primary radiation leakage is very
small (depending of course on beam energy-Proimos was using 2 MeV radiation

-.,,,, L \ c’ I
Eye level
cross section
Figure 2.12. ( a )Synchronous protection of the eyes using two lead blocks.
The eyes are always in the shadow of the blocks whatever the beam
orientation. In this drawing the source is stationary ana‘ the patient rotates
about a vertical axis through C and the blocks are on a co-rotating plate
in the jield with an additional arrangement (shown in ( b ) on the alternative
graviry-support mechanism for stationary patient and rotating source) to keep
them parallel to the rays at all rotational positions of the plate. (From
Proimns (1961,1963).)
from a Van de Graff generator in 1961). The limit to the protection afforded to the
shielded region in ‘total eclipse’ is determined by the scattered radiation. Figure
2.12 shows an example of how this technique was used to shield the eyes.
Proimos described two possible arrangements for treatment. Either the source
remained stationary, the patient rotated about a vertical axis and the shields rotated
on a plate with an arrangement to keep the lead blocks parallel to the rays (as
in the example shown in figure 2.12(a) and as orginally proposed by Trump et
a1 (1961), or alternatively the same result was achieved by rotating the gantry
about a horizontal stationary patient with the shielding suspended by gravity with
synchronous block tuming (figure 2.12(b)). A form of this second arrangement
is also illustrated in figure 2.13, where an absorber shaped to the geometry of the

/ I--- \
Figure 2.13. Synchronous protection of the spinal cord. The absorber,

shaped to the cord, hangs simply under gravity. As the absorber has a circular
cross section, it does not need to be rotated. As the source rotates, the spinal
chord is always in the shadow of the absorber. (From Proimos (1961).)
spinal cord is being used to shield it.

Proimos (1961) also used this rotational technique for shaping fields in an
attempt to conform the high-dose region to tumours with irregular outlines. Figure
2.14 illustrates this. As the patient rotates anticlockwise about the axis C, a
pair of absorbing ellipses separately co-rotate to present a beam whose cross
section depends on the orientation. The high-dose area is the ellipse shown
in the patient section. The figure diagrammatically shows the patient rotating
and the field-shaping ellipses co-rotating relative to a stationary x-ray beam; in
practice the beam would rotate about a stationary patient and once again the
appropriate arrangements would be needed for the field-shapers, i.e. the two
elliptical absorbers would require to be rotated in the opposite sense to the gantry.
As for protectors, this can be arranged by letting them hang under gravity, as
shown in figure 2.15, providing a variable beam width. By arranging that the
absorbers create an aperture matched to the projection of the target region in the
direction of the beam, more complicated high-dose regions could be achieved. (In
a sense this is not unlike the modem shaping of the leaves of a multileaf collimator
to the beam’s-eye-view of the target region-see chapter 5.) The shield A in figure
2.14 would hang also under gravity as described above. If, instead of a circular
rod, this were a block, there would also need to be a separate mechanism to rotate
the block, aligning it with the rays. Proimos presented mathematical arguments to

LATERAL POSITION
ANTERIOR POSITION
Figure 2.14. Treatment of an elliptical area in the pelvis by synchronous

field shaping. Note that the circular absorber at A protects the spinal
cord at B . The dose distribution obtained is shown below the irradiation
arrangement. (From Proimos (1961).)
show that the dosimetry was independent of the position of the isocentre relative
to the high-dose area.
In his earliest paper on synchronous megavoltage therapy, Proimos (1960)
had proved geometrically that if the pair of field-shaping absorbers were to co-
rotate about an axis, not at their centre of gravity but offset by some fixed
distance, then the corresponding region in the patient which would always receive
radiation would be offset from the isocentre by the corresponding magnified offset.
He described this as like a searchlight beam following the course of a circling
aeroplane. Then it followed immediately that if disks of different size and offset
were all set up on a single axis so they could be simultaneously rotated, the high-
dose volume would develop a corresponding shape which could be significantly

I
I
Figure 2.15. Showing how the use of apair of shaped blocks rotating under
gravity can be used to shape a rotating x-ray field and thus conform the
high-dose region to an irregularly shaped treatment region. The absorbers
rotate on axes AB, CD and are shown in four orthogonal positions. (From
Proimos (1963).)
different from a right cylinder of constant radius. This is illustrated in figures 2.16
and 2.17. The method was further exploited by Ilfeld et a1 (1 97 1).
Later Proimos (1 969) designed more complicated gravity-assisted absorbers,
cylinders containing a largely unattenuating demagnified tumour model sur-
rounded by lead shot to tailor the high-dose volume to the irregularly shaped target
volume.
Recently a novel arrangement has been engineered by Casebow (1990a)
whereby this gravity assist plus geared rotation is combined with the use of a
circular absorbing pin and two wedge filters whose thinnest ends abut the pin.
This is analogous to the inverse of the filtered backprojection used in computed
tomography imaging (figure 2.18) and was shown by 'end-on' film dosimetry to
give a better dose uniformity in the unshielded region than the use of the pin alone.
Casebow ( 1 990b) has suggested that one might logically consider extending this
treatment technique to make use of a more complex absorber with an irregular
shape and perhaps constructed of materials of differing x-ray absorption. This
could introduce a beam intensity function f (r&,4) with the properties required to
approximate the prescribed dose d (r, 0) via equation (2. lb). Suppose the required
functions f ( r & , 4 )could be obtained by some method (see e.g. section 2.5); then
the absorber would be specified by the attenuation matrix p ( x , y) which is the CT

X-
so
ABSORBING CROSS
, ,MACHINE WINWW
Vertical section of s m k arriagement in front of machine aindaa
Figure 2.16. ( a )How a region not centred on the rotation isocentre can be
irradiated by using a pair of offset corotating field-shaping cylinders. The
patient is assumed to rotate in the fured beam. The patient rotates about C1
and the field shapers corotate about 0 1 and 02.As they do so the beam
is swept like a search-light following the course of the target volume. ( b )
By stacking sets of these field-shaping disks on two corotating shufts, the
high-dose volume can he conformed to a highly asymmetric volume, not
centred on the axis of rotation. (From Proimos (1960).)
reconstruction from f (r$,4). In principle this would provide a very neat solution,
but there may be practical difficulties such as:
0 there may be no consistency in f (r$,4) to give a solution,
0 f ( r # , $ ) may not be properly normalized (recall that in x-ray CT scanning the
area under all projections must be the same),
0 and the solution, if it exists, may not be well sampled (CT typically uses
hundreds of projections).
Most of the papers describing synchronous field shaping and synchronous
blocking were somewhat qualitative and did not directly address the dosimetry.

Figure 2.17. How synchronous absorbersfor field shaping andsynchronous

protection can be arranged to provide a high-dose volume in the continuously
irradiated volume (CIV) and protect the intestine ( I ) , bladder ( B ) and rectum
(R). (From Irfeld et a1 (1971).) (Courtesy of the American Rontgen Ray
Society.)
L
relative position
Figure 2.18. Design of wedge + pin absorbing filter. The filter is mounted
on aplate which hangs under gravity and the absorbers synchronously rotate
by an arrangement of gears. The curves show the transmission profiles,
normalized to block width (a,b, measured at 10 cm depth in water for 1.85
cm and 2.8 cm width blocks and c is a design calculation). (From Casebow
(1990a).)
This was put right by Rawlinson and Cunningham (1972). They noted that
the dose within the shielded region was no2 negligible and that the dose in the

1
9C l 5 cm 4
8
Radial distavce i c m l
12
Figure 2.19. The radial dependence of the dose profile in rotation therapy
with synchronous protection. The absorber is of width 2 H A and the field
of width 2 H B (shown as an inset). The data points are experimental
measurements. The heavy broken curve is the result from the simple theory
given by equations (2.26) to (2.28)showing good agreement. The dashed
curve is the result without the block (open beam). A full 360" rotation was
used, The phantom was 30 cm in diameter. The values for H A and HB are
shown. (From Rawlinson and Cunningham (1972).)
continuously irradiated region was not uniform. They made measurements using a
full 360' degree rotation of a uniform field with width 2H B with a central absorber
of width 2 H A (figure 2.19) irradiating a cylindrical phantom of 30 cm diameter
with the isocentre in the centre of the phantom. Note the approximately 20%dose
to the shielded region, the scatter into the shielded region (on the experimental
curve) and the non-uniformity in the continuously irradiated region (because the
cosine dependence needed (equation (2.21)) was not included).
For this circularly-symmetric situation they computed the expected radial dose
variation by simple trigonometry. From equation (2.lb) , we note that we may
set 8 to any value we choose. The 8-dependence of d and the 4-dependence off
disappear and setting 8 = 7c/2 equation (2.lb)becomes
d (r)= -1
1
2n
2n
f ( r sin#) d#. (2.23)
Recall that the approximation in equation (2. lb), and hence also in equation (2.23).
is that the dose to any point in the dose space is given by backprojecting the beam
element 'connecting to' that point. Put another way, the dose beneath the absorber
is a constant value for each position in the field. Set this to unity for the open field
and to L for below the absorber, i.e.
f(rsin#)=L for(rsin#)<HA (2.24)

and
f(rsin@)=l forHA < ( r s i n @ ) < H B . (2.25)
It follows immediately that
d(r)= L for r < HA. (2.26)
For HA < r -= H B, f = L for a total rotation angle of 4 sin-' (HA/r) and unity
for the remaining part of the 2rc rotation. Hence
d ( r ) = 1 - (' - L)sin-' (HA'r)

for HA < r < HB. (2.27)
XI2
For r > H B, f = L for a total rotation angle of 4sin-' (HA/r) and unity for a
total rotation angle of 4 sin-' ( H B / r ) - 4 sin-' (HA/r) . Hence
sin-' ( H B / r ) (1 - L ) sin-' (HA/r)

d (r) = - forr > HB. (2.28)
El2 n/2
Rawlinson and Cunningham plotted equations (2.26) to (2.28) (see figure (2.19))
and found that despite the very simple approximations (such as ignoring photon
attenuation and scatter) in this model there was good agreement with the
experimental data points. The poor uniformity in the continuously irradiated zone
is the result of rotation wherein for part of the time this region 'sees' the absorber.
They suggested using a wedge with its thinnest end abutting the absorber as also
suggested by Casebow (1990a) and discussed above.
By a similar geometrical construction as used to get equations (2.26) to (2.28)
it can easily be shown (see e.g. Brahme 1988 and Boyer et a1 1991) that when
a uniform beam of width 2a rotates about 360° the primary dose for any radius
r > a is given by
d (r) = - sin-'
n ' ($1 (2.29)
within the p = 0 approximation. This non-zero dose is unwanted (Boyer et a1

1991 call it the 'transit dose skirt') but unavoidable if the region r < a is to
be approximately uniform. The radius r50 at which the dose becomes 50% of
the dose in the continuously irradiated area is a a , which is very similar to the
value found from the theory in section 2.3 when attenuation (but not scatter) was
included and the beam profile had a spatially varying intensity (see equations (2.9)
and (2.22a, b))to obtain an exactly uniform dose inside the radius rl .
In this section an account of synchronous field shaping and protection in
rotation therapy has highlighted the contemporary interest in those techniques. In
passing we should also note that altemative techniques called 'tracking therapy'
for conformation therapy were being developed in London (Davy 1985, Brace
1985). Work started as early as 1959 (Green 1959) at the Royal Northern Hospital,
progressed in several phases (Green 1965) and the Mark 2 machine was eventually

Methods for 2 0 and 3 0 optimization 87
sited at the Royal Free Hospital. One of the prime movers, W A Jennings,
has recently provided a historical overview of these developments (Jennings
1985). Green (1965) wrote ‘...the basis of attaining the principle of restricting
the radiation to the diseased area while sparing the normal tissues around appears
to be well within sight.....’.
2.5. METHODS FOR 2D AND 3D OPTIMIZATION
2.5.1. Analytic and iterative deconvolution theory of 2 0 inverse computed

tomography
In section 2.3 we showed how it was possible to compute a point-irradiation
distribution corresponding to the rotation of a very thin pencil beam (of width
2rl) around a full 2rr rotation. The dose distribution was uniform inside a circle
of radius r1 and thereafter fell off monotonically. It is this monotonic decrease
which rotation therapy attempts to exploit. To emphasize this we now show that
using a parallel-opposed pair of fields only would give much more dose to the
region traversed by the fields. Brahme (1988) described the central-axis depth
dose distribution in a single photon beam by the function
(2.30)
where the first exponential describes the fall-off in absorbed primary dose and
the second exponential describes the dose build-up due to secondary electrons, /.Le
represents the attenuation coefficient of the secondary electrons and U is a measure
of the electron contamination in the beam. Now if two such beams are parallel
opposed, directed to a cylinder of radius R , the dose normalized to DOat the centre
of the cylinder and expressed in terms of the distance r from the centre is dpo ( r ) ,
where
+
PO ( r ) = DoEexp ( - p ( R + r>>- v exp ( - p e ( R + r ) ) exp ( - p (R - r ) )
- exp ( - F e ( R - r))1/[2 (exp (-pR) - vexp (-peR))I
(2.3 1)
and PO refers to ‘parallel opposed’. This expression reduces, using hyperbolic
functions, to
cosh p e r - cosh p r
(l/v) exp ( b e - CL) R ) - 1
). (2.32)
(If electron build-up is ignored, then one may set U = 0 and only the first term
remains namely
dpo ( r ) = Docoshpr (2.33)
which is of course the simple expression from Cormack (1987) (see section 2.1
just above equation (2.1c)). Brahme (1988) has plotted the distribution given by

equation (2.33) on the same diagram as that given by equation (2.9) with (2.22a),
(2.226) for the point irradiation distribution corresponding to the rotation of a very
thin pencil beam (of width 2rl) (where one reads DOfor D in equation (2.22~)).
These two curves are shown in figure 2.20. The trade-off is now clear:
rotation therapy delivers dose everywhere in the slice but concentrates it at the
centre of rotation;
paraflel-opposedpairs give dose only in the beams’ path (ignoring scatter), thus
sparing tissue not in the ‘line of sight’ but giving a much less ‘concentrated’
dose.
Let us now return to considering the point-irradiation distribution. As shown,
this can be calculated with certain approximations. Alternatively it could be
measured experimentally. It could also be deduced from the point-spread function
as follows, using the notation of Brahme (1987, 1988). Define the point-spread
function h (r) as the ratio of the mean energy imparted per unit volume at the
point r by the photon interacting at the origin r = 0 (this could be computed by
Monte Carlo methods for example). For a pencil beam directed down a z-axis, the
pencil-beam dose distribution p (r) is given by
p (r) =
s exp (-pz) h (r - z) dz.
From the pencil-beam distribution the cylindrical convergent point-irradiation

(2.34)
distribution may be computed by
df (r) = f p (r)dl/2xr (2.35)
where 1 refers to distance around the contour.

One can also imagine a point-irradiation distribution dspoke from a finite number
of pencil beams, equispaced in 2rr, which would have the form of a spoke
pattern centred on the rotation axis (Lind and Kallman 1990). From here on we
shall use the symbol d (r) to represent the rotational point-irradiation distribution
howsoeverformed, with it implicit that this is appropriately selected by one of the
methods given above. It is the ‘fingerprint of the irradiation’ (Brahme 1992).
The question arises of the spatial invariance of the function d (r). Brahme
(1988) argues that the functions are only weakly spatially variant and, at least to a
first approximation, may be considered spatially invariant. True spatial invariance
would require an infinite source-to -skin distance (SSD). As the SSD is reduced, the
divergence of the beam increases and the approximation becomes less appropriate.
In the experimental verification of the inverse tomography algorithm of Lind and
Kallman (1990) (see later) 2 m SSDs were used to minimize the problem of the
positional sensitivity of d (r).
Brahme then argues that from the superposition principle the dose D (r) at any
vector r in a cross section of the patient may be obtained from the density @ (r)

Methods for 2D and 3D optimization 89
Figure 2.20. Showing the difference between the point irradiation distribu-
tion from a full 2n rotation (dotted) and the central axis distribution (solid)
for a parallel-opposed pair of pencil beams. The dotted curve is for a slit
beam of width 2u. as shown in the right-hand side of the lower part of the
Bgure. (Adaptedflom Brahme (1988).)
of point irradiations (sometimes known as the kernel density (Lind 1990))via the
convolution integral
(2.36)
where r’ is the centre coordinate of each point irradiation. Equation (2.36) is

a straightforward convolution equation. It is the space-invariant form of the
Fredholm equation of the first kind:
D (r) = 11 (r’)d (r,r’)d2r’ (2.37)
where d (r,r’) is an elementary dose kernel representing the dose at r from an

elementary irradiation convergent at r’. From this we see that the irradiation
density function @ (r) is given implicitly by an integral equation containing D (r)
and d (r).Equation (2.36) is the forwardplanningproblem, i.e. knowing the two
quantities on the right-hand side, the left-hand side follows. The inverse problem
is alternatively that of inverting equation (2.36) to obtain this irradiation density
function. In principle this can be achieved by a deconvolution, since, if we cast
equation (2.36) in Fourier space, this becomes
D (s) = @ (s) d (s) (2.38)

where the Roman font represents the Fourier transform of the corresponding italic-
font character and s is the Fourier-space variable inverse to r. Equation (2.38) can
then be inverted to give
(2.39)
The term q (s) is a low-pass filter function which should be designed to ensure
that the deconvolution does not become ill-conditioned when the function d (s)
has zeros.
The final step in calculating beam profiles is now to make 'inverse
backprojections' (or forward projections) of the density function @ (r) weighted
by the positive exponential factor representing attenuation of the beam, i.e.
where the lower case 4 represents the angle of the projection, as in section 2.1, and
r6 represents the position within that beam; the vector r labels the position in the
slice where the intensity function is @ (r)and z (r)is the depth of the point r in the
direction of the projection of the line integral. If the slice is very inhomogeneous,
the single exponential should be replaced by a ray-trace through the tissues of
differing attenuation (see also Lind 1990). If the point-irradiation distribution
were a simple spoke pattern, then the projections should only be made in the same
directions as the angles of the spokes. Of course one cannot have negative beam
weights, so where these arise the beam-weights must be set to zero. In this sense
this calculation is only an approximation. The beam profile in equation (2.40)
must be computed by ray-tracing along fan-lines connecting the point r to the x-
ray source point.
There is an altemative way (Lind and Kallman 1990) to compute the resultant
dose distribution D (r)from the beam profiles f (r#, 4). Instead of using equation
(2.36), which uses the density of point irradiations and the point-irradiation
distribution, one can backproject a function f* (r$, 4) derived from the beam
profiles via
D (r) =
#
1 f* (r;, 4) exp (-/U(r))dr; (2.41)
where f* is f convolved with the line spread function (which could be obtained
from a Monte Carlo calculation). Equation (2.41) provides a direct method of dose
calculation completely analogous to the usual way of computing dose distributions
from superposed open or wedged fields. The only difference here is that the beam
profiles are highly non-uniform.
At least in principle, then, the spatial profile of beams at differing orientations
may be computed and then delivered either using a multileaf collimator or sets
of fixed compensators or scanning photon pencil beams. Perhaps this is easier to

Methodsfor 2D and 3D optimization 91
imagine than to achieve. Convery and Rosenbloom (1992) have recently provided
a potential solution in terms of moving jaws (or multileaf collimator leaves) with
a time-varying velocity. The above theory is due to Brahme (1988). Similar
arguments have been advanced by Mackie et a1 (1985).
Instead of using the deconvolution equation (2.39) to invert the integral in
equation (2.36), an iterative method may altematively be used. Lind and Kallman
(1990) have used the iterative scheme:
@o (r) = a D (r) (2.42)
@k+l (r) = c (@k (r) a (D (r) - @k (r)@ d

+ w)) (2.43)
where the subscript k refers to the kth iteration and @ means convolution. They
use typically 20-30 iterations. C is a positivity constraint operator and a controls
the speed of convergence and deals with the units and dimensions. Lind (1990)
writes the same equation in matrix notation as
where the symbols are now read as vectors (a,D)and matrix operator (d). It tums
out that this solution never gives underdosage. Lind (1990) also give the iterative
scheme
@k+l = c ( @ k +
a (d'D - d'd@k)) (2.45)
which is the result of minimizing in a least-squares sense. This is formally
identical to iterative methods to remove blurring from medical images (see e.g.
Webb 1988).
Lind and Kallman (1990) investigated the first iterative method for two model
clinical problems. The first attempted to obtain a uniform dose inside a 12 cm
square within a 25 cm diameter cylinder using just three beams at 120" to each
other. For this case the point-irradiation distribution (dose kemel) was of the
dspke type described earlier with three spokes and was generated by Monte Carlo
methods. Figure 2.2 1( a ) shows the desired dose distribution, figure 2.21'3) the
point-irradiation density and figure 2.21(c) the result of applying equations (2.42)
and (2.43). Figure 2.21(d) shows the resulting beam profiles. The delivered dose
distributions were then computed by equation (2.41) and measured by constructing
compensators to give the required beam profiles (thickness related to the log of
the beam profile divided by the linear attenuation coefficient of the compensator
material). There was good agreement, but, needless to say, a perfect square cannot
be achieved with this small number of fields.
The second dose distribution was a cylindrically-symmetrichigh-dose volume
with two dose levels (figure 2.22(a)). The point irradiation density corresponded
to a full 360" rotation (figure 2.22(b)). The density of point irradiations resulting
from iterative equations (2.42) and (2.43) is shown in figure 2.22(c). In this case
all profiles are of course the same. Again the delivered dose distribution was

Figure 2.21. ( a ) The desired dose distribution, a uniform dose inside a

square of side 12 cm within a 25 cm diameter circle (not shown). (b) The
point irradiation distribution from just three equally spaced 2 MeV beams.
( c ) The density of point irradiation from the iterative calculation. ( d ) The
three beam profiles obtained by ray-tracing throught the irradiation densities
in c. (From Lind and Kallman ( I 990).)
computed by equation (2.4 1 ) and measured experimentally with film dosimetry

with good agreement.
Remembering the theory developed in section (2.3) for this circularly-
symmetric problem there exists an analytic solution (equation ( 2 . 2 1 ~ ) )Figure
.
2.23 shows a comparison of the analytic solution obtained from this equation with
the absorbed dose profile through the centre of the target volume calculated from
the iterative algorithm (2.42) and (2.43) and equation (2.41).

Figure 2.22. ( a ) The desired circularly-symmetric dose distribution, a

two-level uniform dose within a 12 cm circle. ( b ) The point irradiation
distribution from a continuous 360’ 2 MeV irradiation. ( c ) The density of
point irradiation from the iterative calculation. (From Lind and Kallman
(1990).)
Note the characteristic ‘ears’ to the density of point irradiations near the
sharp boundaries of the dose distributions (figures 2.21(c), 2.22(c)). This is to
compensate for the absence of in-scatter from surrounding volumes.
Lind and Brahme (1992) have discussed a number of different geometrical dose
kernels d ( r ) appropriate to other treatment geometries which may also be used
in the same way with the convolution equation (2.36) as the basis of treatment
optimization.
The theory of iterative deconvolution has recently been extended by Kdlman ef

Figure 2.23. Comparison of the absorbed radial dose profile computed by

the iterative method (solid line) and the analytic method (dotted line). (From
Lind and Katlman ( 1 990)J
a1 (1992) to include modelling the biological response of normal and pathological

tissues. The equations were reformulated to maximize the probability of
uncomplicated tumour control taking into account the distribution of clonogenic
cell density in the target and the ‘organization’ of the tissues in terms of the
serial and parallel nature of the functional subunits. Kdlman et a1 (1992) applied
both this biologically based optimization and the ‘dose-only-based’ iterative
optimization to the same model problem, with the result that the former gave
a small increase in the probability of uncomplicated tumour control. It was
also shown that the resulting 1D beam profiles were smoother with the former
method. The biological optimization does not require the target volume to have
homogeneous dose, but tends to reduce the dose to parts of the target close to
organs-at-risk and increase the dose to compensateelsewhere in the target volume.
Bortfeld er a1 (1990a, 1992) also develop the theme of computing beam profiles
by the techniques of inverse computed tomography for situations with only a small
(and odd, such as 7 or 9) number of beams. They describe the function f (r$, 4) as
an intensity modulationfunction or IMF. The method is similar to that of Brahme
(1988), but not the same. For instance they do not refer to forming projections of
the density of point irradiations, as in equation (2.40) but instead refer to a first
step of forward projecting the dose prescription itself, i.e. although they do not
explicitly write the equation, they perform
f (r$>4) = 1line
D (r>exp (CLZ (r))dz. (2.46)
This is reasonable given the first-order approximation from Lind and Kallman’s
method that @O (r) = a D (r)(equation (2.42)). However when they compute the
backprojection from these first-order profiles
D (r) = f (r$?9)exp (-PZ (r)) (2.47)
4
the resultant dose distributions were not as tightly conforming as required. To

overcome this, they high-pass filtered f (r$,@)before the backprojection (setting

Image Reconstruction Conformation Radiotherapy

(Filtered Backprojection) (Filtered Projection)
Density Distribution Set of Prescribed

of the Tissue 2D Dose Distributions
1
onto Lines
I Backprojection
+
Set of 2D Slice Images
that are given by the
Dose Distribution
in the Tissue
Figure 2.24. Illustrates the formal similarity between image reconstruction

by the CT technique of filtered backprojection and the filtered-projection
method of Borgeld for conformation radiotherapy. The actions in the
lighter boxes are the practical tasks. The actions in the darker boxes
are the backprojection or formation 05 and the actions on, the projections
respectively. (From Bortfeld et al(I990).)
any negative values to zero). This improved the degree of conformation. It would
appear that this filtering is needed simply because Bortfeld et a1 (1990a,1992)
did not choose to forward project (as in equation (2.40)) the densityfunction 0
which would have arisen from a deconvolution of the point-irradiation distribution
(equation (2.36)). They provide a diagram (figure 2.24) which clarifies this point
whilst discussing the analogy with computed tomography.
Bortfeld et al (1990a,1992) then adjust the beam elements by an iterative
method, starting from the filtered projections, which minimizes a ‘cost function’
based on the RMS difference between the prescription and the running estimate of
the dose distribution. This is similar to the simultaneous iterative reconstruction
technique in medical imaging tomography. After some 10 iterations for a test
problem the result is a substantial improvement.
We have seen that the point-spread function (PSF) is central to these
computations. So now we introduce a lengthy interlude on convolution techniques
for dosimetry. Central to these is the use of the PSF, usually computed by Monte
Carlo methods. Readers wanting to stay with the central theme of optimization
should proceed now to section 2.5.4.

2.5.2. A note on convolution dosimetry based on terma

Provided the pattem of energy spread around a point of primary photon interaction
is spatially invariant, convolution methods may be used in dosimetry. Such
methods are very modem and have not yet found their way to all planning
computers used for routine patient dosimetry, which at present still use some
of the methods reviewed over ten years ago by Cunningham (1982). Equation
(2.34) was a specific example of a convolution for calculating a pencil-beam dose
distribution. Strictly, spatial invariance will only hold for uniform homogeneous
phantoms. Ahnesjo er a1 (1 987a) defined the point-spread function (sometimes
called an energy deposition kernel (Ahnesja 1989)) as the fractional mean energy
imparted in a small volume d3r at vector position r when a photon interacts at the
origin of coordinates, i.e.
(2.48)
The dose is the convolution of this function with the terma at energy E , the total
energy released by primary photon interactions per unit mass TE (r). Terma has
dimensions energy per unit mass. The terma at energy E , TE (r),is given in terms
of theflueme at energy E , r E (r) (dimensions photons per unit area), via:
where p / p ( E , r) is the mass attenuation coefficient of the primary photons of

energy E at point r (dimensions area per unit mass). The artenuation rather than
the absorption coefficient is used because the rei" includes energy carried by
both photons and electrons (compare with section 2.5.3, equation (2.62)). The
energy carried by electrons alone would be the traditional kerma. (In other papers
(e.g. Ahnesja 1989, Ahnesjo and Trepp 1991) energy is combined with fluence to
give 'energy fluence'.) From equation (2.49) the dose at energy E is
D E (r) = 1 11 TE (s) h ( E , r - s) d3s (2.50)
where the specific dependence on energy of the point-spread function has been
inserted. The fluence at energy E is given by (figure 2.25)
(2.5 1)
where p (I) is the density. Equation (2.51) is simply the inverse-square law
with exponential attenuation applied to the photon fluence at the surface (ro).
Equation (2.5 1) is rather oversimplified and Mackie et a1 (1987) explain that,
for example, at the field edges the penumbral fluence must be developed by
convolving with a Gaussian point-spread function, which accounts for the finite
x-ray source size. The small penetration through the collimation can also

Figure 2.25. Showing the relationship between thefluence at distance rand

ro from the source (see equation (2.51)).
be included. Blocks and wedges can be similarly handled by building in a

modification to FE (ro).
The total (rather than differential) terma is simply (from equation (2.49))
(2.52)
and the corresponding total dose from photons of all energies in the incident
spectrum is (from equation (2.50))
D (r) = //// TE ( s )h ( E , r - s) d3s d E . (2.53)
Looking at how Brahme’s equation (2.34) fits into this scheme we see that
Brahme combined the energy into the point-spread function (dimensions energy
per unit volume; Ahnesja et a1 (1987a) had simply per unit volume) and took the
energy out of the other term under the integral.
Ahnesjd et a1 (1987a) computed the point spread functions by Monte Carlo
methods, properly accounding for all leptons, and their paper indicated they
will supply these distributions to other workers. These distributions can also be
represented by analytic expressions (see below). They used them to compute
depth-dose curves with enormous accuracy. For example, typically 500 000
histories were used per point-spread function, taking about 80 h on a VAX 750.
Using 200 OOO dose points in equation (2.53) is thus equivalent to a Monte Carlo
run taking 1800 years of VAX 750 time!
The above convolution equations strictly only hold for a space-invariant point-
spread function. Ahnesja (1987) considered this further. Firstly, the point-spread
function can only be spatially invariant for an infinite medium and an error would
arise if the kernel was used near to the boundary of a medium. Ahnesja developed
a ‘worst-case’kernel, the spatial average of a point-spread function corresponding
to the primary interaction site forced to the centre of the entry and exit surfaces of a
large phantom. This was compared with the point-spread function developed from
an interaction point in the centre of the phantom by using both to compute depth-
dose curves. The mean difference between the two was 0.2% and the maximum

difference was 1.6% (at small depths for large fields). Thus spatial invariance was
considered a reasonable approximation.
Secondly, the convolution equation assumes a kernel aligned to the main axis
of the incident field. Ahnesja (1987) worked out that for diverging beams the
error in this approximation increased with depth and with decreasing source-to-
skin distance. For example, the error was 3% at 25 cm depth for SSD = 100 cm.
Mackie et aI(l985, 1987) explain that if the beam is tilted relative to the patient,
the dose kernels must tilt correspondingly.
Ahnesjd (1 987) found almost no change in kemel due to beam hardening. If a
kernel h (r) is defined from the kemel with energy dependence E , h ( E ,r) via a
weighting by the fluence
(2.54)
then the fourth integral in equation (2.53) can be removed to give
D (r) = /// T (s) h (r - s) d3s. (2.55)
This considerably shortens computation times. Ahnesj@ (1989) included a

multiplicative ‘fudge factor’ in equation (2.55) as the ratio of dose calculated for
homogeneous water by equation (2.53) to that from equation (2.55).
Mackie et a1 (1987) explain that it is simply too time consuming to work out
dosimetry for spectral beams by convolution with monoenergetic kemels and
subsequently weighting the results by the amplitude of the spectral components.
Computing a dose kernel weighted by the spectral components is likely to be the
only practical solution to the problem. Additionally, a potential problem arises
because the mass attenuation coefficient for a spectral beam depends on depth
into the scattering medium because of beam hardening. In principle, a spectral-
averaged mass attenuation coefficient can be computed by a formula given by
Mackie et a1 (1987). However, this is implicit in equation (2.52) above, where the
differential terma (which requires the same spectral weighting) is integrated to get
the total terma.
Ahnesjd and Mackie (1987b) found that the monoenergetic point-spread
function h ( E , r)could be represented accurately in the scatter zone by the analytic
function B exp (-b I r I) / I r I*, where B and b depend on energy and the angle
8 of r to the direction of the impinging photons. Ahnesja (1989) found that the
polyenergetic kemels h (r) (equation (2.54)) could be represented accurately by
the analytic function in water:
h ( I , 8 ) = ( A e exp (-aer) + BOexp (-her)) / r 2 (2.56)
where A H ,a@,Be and bo are functions of the accelerating potential and of the angle
8 from the measurement point r to the direction of the impinging photons (figures
2.26 and 2.27). Large tables were provided by Ahnesjd (1989) for a range of 6 at

Figure 2.26. A gamma ray interacts at a point. The point-spreadfunction

is determined by the distance rand the angle 0.
FRACTIONAL ENERGY IMPARTED
Lateral distance tcml
Figure 2.27. Showing the energy deposition kernel for 6 MV photons in

water. The thin hatched lines are calculated with Monte Carlo methods,
whereas the thick solid lines are the corresponding analytical fits using
equation (2.56).(From Ahnesj# (1989J.j
different accelerating potentials. The first term is mainly primary dose; the second
is mainly scatter.
By deconvolving the point-spread function (convolution kernel) from a
measurement of dose at depth (using for example equation (2.55)) the energy
fluence may be computed (Ahnesjo 1990). Ahnesjo and Trepp (1991) amplified
this theme.

Clearly, lack of homogeneity in the tissues must render the point-spread

function space variant. Some workers have overcome this problem by scaling
the kemel in terms of radiological distance from an interaction site to the site of
dose measurement (e.g. Ahnesjo 1987, 1989, Mackie et a1 1985, 1990a,b, 1991,
Nilsson and Knoos 1991). Equation (2.50) can be cast into the form
DE (r) = 111 TE ( r - s) h ( E , s) d3s (2.57)
which becomes
DE (r)= 11 1 p / p ( E , r - s) E r E (r - s) h ( E , s,U )1s12dlsldC2. (2.58)
When the 3D integral is explicitly broken into parts, the terma is written in terms
of the fluence and w is introduced into the argument of the point-spread function
to indicate the direction s from the interaction site. Equation (2.58) holds for
homogeneous media. The form in an inhomogeneous medium follows by simply
scaling the vectors into the radiological length p l (s) namely
This is the form of the equation given by Mackie et a1 (1990a).

The use of scaled kemels removes the opportunity to perform convolutions
in Fourier space. Zhu and Boyer (1990) have presented an altemative way to
compute the dose in 3D for inhomogeneous materials. The kernels remain spatially
invariant but thefluence is computed by ray tracing through the 3D CT matrix
of inhomogeneous densities. This does not jeopardize the possibility of using
Fourier space with its attendant speed advantages. Zhu and Boyer (1990) show
the method compares well with measurements for fairly simple inhomogeneous
phantoms. Obviously it fails in the region of large interfaces, where there is no
electron equilibrium.
When the kemel is space invariant the convolution equations given above can
be implemented in Fourier space by more time-efficient multiplications (Boyer
and Mok 1985, Boyer et a1 1988, Field and Battista 1987, Mackie et a1 1985,
1990a). Typical speed-up factors are given in section 2.5.3.1. Altematively,
Bortfeld et a1 (1990b) describe how the 3D convolutions can be separated into 2D
and 1D convolutions. When, on the other hand, the kemel is not space invariant,
these methods are unavailable (Ahnesjo 1989). The real-space calculation can
be very time consuming and Mackie et a1 (1990a,b) discuss some methods of
speeding up the computation and report the method in use in the AB Helax
treatment planning system. Zhu et a1 (1992) have used the method of convolution
dosimetry for irregular fields defined by a multileaf collimator and showed that the
results matched experimental determination to an accuracy of some 3 mm. These

DIFFERENTIAL PENCIL BEAM DEFINITION
i’
Beam
T I
Poin! of
Collision
‘Pa I
Figure 2.28. The definition of the differential pencil beam. The direrential
pencil-beam dose distribution is the dose distribution relative to the point of
first collision of the primary photons. It is so-called because its line integral
gives the dose due to a pencil beam. (From Mohan et a1 (1986).)
calculations were performed with a voxel size of 5 mm and took some 30 min per
field (on a VAXstation 3500) for a 64 x 64 x 128 dose matrix.
2.5.2.1. Convolution dosimetry by the differential pencil-beam method. Mo-

han et a1 (1986) introduced the differential pencil beam method of convolution
photon dose calculation, The formalism is given here and it can be seen that it has
similarities with the method based on the concept of terma.
The following definitions were introduced. A pencil beam is a parallel
monoenergetic photon beam with infinitessimal cross-sectionalarea (figure 2.28).
A differential pencil beam (DPB)at a point (such as Q) is the fraction of the pencil-
beam photons which have their first interaction in a small volume surrounding that
point. The collision density is the number of first collisions per unit volume in
a small volume surrounding the point of first collision. The differential pencil-
beam dose distribution is the dose distribution relative to the position of the first
collision, per unit collision density, for such a monoenergetic pencil beam in an
infinite homogeneous medium of unit density. The dose distribution at any point
is then computed as the convolution of the collision density and the differential
pencil-beam dose distribution.
Following Mohan et a1 (1986), the collision density at point Q is
rQ( E )p ( E )exp (-p ( E )t ~, where
) rQ( E ) is the fluence at Q of the unattenu-
ated photons of energy E , p is the linear attenuation coefficient for these photons
and tQ is the depth of point Q in water. The differential pencil beam (for dose de-
posited at point P from photons having their first interaction at point Q) is written
D P B (rpQ,8, E ) . Hence the dose at point P for photons of energy E is given by
DP,=
J rQ ( E ) p ( E ) exp ( - p ( E )tQ) D p B (rpQ,8 , E ) at,. (2.60)
Note the collision density has dimensions [ L - 3 ] and the DPB has dimensions
[dose x L 3 ] .

If we compare equations (2.60) and (2.50) we see that the first quantity under
the integral is tenna multiplied by ( p / E ) and the second quantity, the DPB, is the
dimensionless point-spread function h multiplied by ( E l p ) . So we see that the
formalism introduced by Ahnesjo et a1 (1987a) can be related to that of Mohan et
a1 (1986) when the beam is considered to be a pencil.
Equation (2.60) shows that the dose can be specified by a convolution of the
collision density with the DPB dose distribution. The latter is obtained by Monte
Carlo calculations. Mohan et a1 (1986) show a large number of examples. These
DPB dose distributions include the effects of both the transport of electrons and
photons and include all the phenomena contributing to the dose for a homogeneous
medium.
DPB dose distributions were computed for ‘water-like’ material with relative
electron density 0.4, and when these were overlaid on to the corresponding water
DPB dose distributions with distance scaled by 0.4 they were found to almost
coincide. From this it was concluded that it is the material along the path from
the first collision to the point of dose computation which mainly determines the
effect of different material on dose transport. This suggested the means to account
for tissue inhomogeneities:
0 first compute the collision density taking into account the non-uniform photon
attenuation between the source point and the point of first collision (using the
average linear attenuation coefficient along this path in the exponential term in
equation (2.60));
0 weight the photon fluence by the linear attenuation coefficient in the voxel at
which the first scatter occurs to account for secondary electron and photon
production, which is in direct proportion (because attenuation coefficient scales
with electron density if Compton processes dominate);
0 scale the path between the point of first scatter and the dose computation point
by an effective density.
This recipe requires modification for materials whose composition differs from
water, such as bone. Mohan et a1 (1986) suggest the first step should instead use a
ratio of total linear attenuation coefficient of bone and water rather than a ratio of
electron density to scale the attenuation. The third step should also use an effective
density when scaling the argument of the DPB kemel. The effects of different
composition will matter more for higher-energy photons when pair production is
a significant interaction. At 6oCoenergies the simple scaling by electron density
suffices.
This method is not expected to work in the build-up region since this is not part
of the DPB model. However, it can compute the dose in the presence of wedges,
beam blocks and collimators, provided the effects of these can be modelled into
the fluence.
In section 2.5.3 we shall see that inhomogeneities can be accounted for in a
convolution model based on kerma. As here, the fluence is scaled accurately and
weighted by electron density to account for changed production. The change in

the kerma kemels is accounted for analytically.
2.5.3. Convolution dosimetry by analytic methods based on kerma

We have seen how the dose to a point can be related to the terma convolved with
the point-spread function or kemel which includes the transport of all energy
including that carried away by secondary photons. Since dose and terma have
the same dimensions, this convolution kemel is dimensionless, The kemel was
evaluated by Monte Carlo methods by Ahnesjo. Provided the kemel is space
invariant, the convolution could be evaluated by Fourier techniques.
There is an alternative way of looking at convolution dosimetry without the
concept of terma and instead using the more familiar concept of kerma. Kerma
is the kinetic energy released by primary photon interactions per unit mass and
therefore includes only that energy imparted to charged particles such as electrons.
It specifically excludes the energy carried away by Compton-scattered secondary
photons and other higher-order scattering. It is not enough to calculate therefore
the kerma deposited by the primary fluence alone, since the total dose is the result
of kerma deposited by primary, secondary and higher-order multiply scattered
photon fluences. All these must be included in the dose calculation. With this in
mind it might be thought that it would not be possible to reduce the calculation to
a convolution of primary fluence and kerma, but Boyer and Mok (1985) elegantly
showed how this could be done and the following analysis stems from their work.
Some assumptions need to be introduced for this to be possible. Assume an
infinite homogeneous medium into which the photons enter. Assume all incident
photons have the same energy. Energy will not appear as a specific dependence
in what follows, but is implicitly present. If the incident beam has a spectrum
of energies, the calculations must be performed for each energy and the result
subsequently weighted by the spectral intensities. For example, Boyer et a1
(1989) show how a linac spectrum can be represented as five discrete energies;
the fluence calculated from Monte Carlo (EGS4) and five separate convolutions
can be summed to give the total dose distribution.
As in section 2.5.2, the primary photon fluence is given by
rp(r)[photon ~ m - ~= ]rP(ro) [photon ~ m - ):~( ]

2
(2.61)
x exp (- lor(0
P
p (0 d)
where p (I) is the density. Equation (2.61) is simply the inverse-square law with
exponential attenuation applied to the photon fluence at the surface rp(ro).ro is
+
the SSD and r is SSD d , where d is the distance into the phantom. The subscript
is to indicate the primary fluence.
The kerma deposited by this primary fluence is

Figure 2.29. Shows a gamma ray entering tissue (surface shown hatched)
and having a first interaction at a point A. Electron transport ensures the
dose is delivered to a region surrounding A. The first-scattered photon travels
to point B where it undergoes a second scatter. In the analytic model it is
assumed the dose from the second scatter is deposited locally at B. In practice,
of course, there is a further small component of electron transport. Multiple
scatter follows thereafter.
K, (r) [cGy] = rp(r) [photon cm-2 ] -

Pen
(EO)[Cm2g-’1
P (2.62)
x EO [MeV] c [cGy(MeV g-’)-’]
where Pen/’P (Eo) is the mass energy absorption (N.B. not attenuation)coefficient
,
at energy Eo. The subscript on this kerma also serves to indicate that it is the
kerma from the primary fluence. Later we shall meet secondary and multiple-
scatter kerma. The factor c = 1.602 x converts from MeV g-’ to cGy. Now
this energy is not of course delivered solely at the interaction site but is carried out
to surrounding points by the electron transport (figure 2.29). This energy transport
may be represented by a dimensionless function f (r) which may be computed
by Monte Carlo methods. Including this transport, the primary dose at r, D, (r)
becomes
D, (r) = 111rp (r’) k!l(EO)Eocf (r - r’)dV’.

P
(2.63)
This is a convolution equation of the primary fluence with what we may call a
kernel for electron transport ke ( r ) ,.i.e
D, (r) = 111rp (r’)k, (r - r’) dV’ (2.64)
(2.65)
Notice that by the stage of writing down equation (2.64) the concept of kerma
cannot be directly identified under the integral. It is shared between the two
functions rp and ke. This differs from the simpler corresponding equation
involving terma instead of kerma (see section 2.5.2; equation (2.50)) where terma

appears explicitly under the integral and is convolved with a dimensionless spread
function including all energy transport (electrons and secondary photons). Note
that ke involves analytic factors but also a function f which cannot easily be
expressed analytically and must be calculated once again by Monte Carlo methods.
This kemel has the dimensions [cGycm2]. We may observe that equation
(2.64) with (2.65) is another convolution (this time fluence with a kerma kemel)
giving dose, analogous to equations (2.50) (based on terma convolved with a point
spread function (PSF)) and (2.60)(based on DPB convolved with collision density).
Both equation (2.64) and the latter involve convolution with either the primary
fluence directly or a quantity proportional to it but (because of the factor ,u in
equation (2.60) operating on the fluence to get the collision density), the DPB
kemel has dimensions [dose x L3], whereas the kerma kemel has dimensions
[dose x L 2 ] . The relationship between the DPB model and the method with the
PSF operating on terma has already been discussed in section 2.5.2.1.
Now we consider the contribution to dose from the first-scattered photons.
These give rise to a scattered fluence r, (r) at a lower energy E,. In tum each
part of the scattered fluence with energy E , produces a monoenergetic kerma K,
due to the first-scattered monoenergetic fluence given by identical formalism to
that for the primary kerma by
K, (r) [cGy] = r, ( r ) [photon cm-’]

Fen
x - ( E , ) [cm2 g-’1 E , [MeV] c [cGy(MeV g-’)-’]
P
(2.66)
where /-Len/P ( E , ) is the mass energy absorption (NB not attenuation) coefficient
at energy E,. The total scattered photon fluence can be written down in terms of
the primary fluence as
Equation (2.67) considers the fluence aniving at r from the primary fluence
at r’, which depends on the Compton differential-scattering cross section, the
inverse-square law for the distance between these two points and the exponential
attenuation (coefficient p’ characteristic of energy &). Pe is the electron density.
The point r receives scatter from a summation over all points r’; so the scattered
fluence from equation (2.67) is no longer monoenergetic. Defining a first-scatter
kerma kernel k, (r),the total kerma due to the first-scatter fluence may be written
K, (r) = / / / rp (r’)k, (r - r’)dV’ (2.68)
with
1 do
ks ( r f ) = (r’)exp ( - p f l r ’ [ ) c A ( E , (r’))E , (r’). (2.69)
(r’Y P

The single prime on p&/p indicates the coefficient is for the first-scatteredphotons
of energy E,. Inside the integral, the fluence at r from r’ is monoenergetic and the
monoenergetic relation (2.69) applies. After integration over all space, the first-
scatter kerma on the LHS of equation (2.69) includes photons scattered to a range
of energies. Now Boyer and Mok (1985) introduce a big assumption, that the
electron transport from the first-scatter kerma can be neglected completely, i.e.
that the dose is delivered exactly where the secondary kerma is released. In this
case the first-scatter dose is exactly equal to the first-scatterkerma and we write
a (r)= Ks (r). (2.70)

For reference we therefore repeat that
D, (r) = 1/ 1rp (r’)k, (r - r’) dV’. (2.7 1)
Thus the first-scatter dose also derives from a convolution of the primary photon
fluence with a kemel.
Now we must tum to the dose which is deposited by higher-order multiple
scattering. Boyer and Mok (1985) argue that after many scatters the remaining
photon fluence (which we call the multiple-scatter photon fluence) rm(r) will be
more or less isotropic. Photons will have ‘forgotten’ their original direction of
incidence. In this case the scattered fluence can be represented by a steady-state
diffusion process described by a Helmholtz equation. It tums out that, if we make
one further assumption (not strictly true, but a simplification)that all the multiply
scattered photons end up with the same effective energy Eavg, a closed solution
exists for the Helmholtz equation. If once again we also assume that the multiply
scattered fluence leads to a multiply scattered kerma K, which is deposited locally
where it is produced (i.e. electron transport is again ignored for these greatly
energy-degraded photon interactions), then dose Dm equals kerma K,. We thus
amve at
Dm (r) = / 11rm (r’)km (r - r’)dV’ (2.72)
where
exp (-lr - r’l/L (Eavg)) Pen”

km (r - r’) = -(Eavg) EavgC. (2.73)
4 n D (Eavg) lr - r’l P
L is a diffusion length and D is a diffusion coefficient. The double primes indicate
the coefficient is for the multiply scattered photons of energy Eavg.
Unfortunately we see that the formula for the multiply scattered dose, although
being a convolution, does not in this form involve the primary fluence. However,
some straightforward but tedious algebra can reduce the expression to this wanted
form, namely
(2.74)

where ks,mis a composite kemel evaluated from k, and k, (see Boyer and Mok
(1985) for details).
Summarizing, equations (2.64),(2.71) and (2.74)give the convolution equations
for the primary dose, first-scattered dose and multiply scattered dose in terms
of the primary photon fluence and three kerma kemels, which can largely be
written down in analytic form. Be aware the kerma kemels do not have the
dimensions of kerma because the fluence remains as a separate term under the
integral. The kemels simply give the ‘extraction factors’ which, operating on the
primary fluence, give dose. The three separate contributions can be added to give
the total dose
D,(r) = ///rp(r’)kt (r - r’)dV’ (2.75)
where
kt (r) + + ks,m (r>.
ke (r) ks (r) (2.76)
This kemel has the dimensions [cGy cm2].
This expression for kt is the three-dimensionalimpulse response of the radiation
dose transport process operating on the primary fluence. If we accept the
limitations which have led to the convolution equation, we may evaluate the dose
by making multiplications in Fourier space rather than the time-consumingthree-
dimensional convolutions in real space. Implementing the fast Fourier transform
(FFT) (Cooley-Tukey algorithm) leads to relatively fast speeds. The Fourier
transform of kt can be thought of as a modulation transferfunction. Indeed all the
ideas from medical imaging carry across on a formal basis when considering the
implementation of equation (2.75). Boyer and Mok present figures showing the
first-scatterkemel ks (forward peaked) and the isotropic multiple-scatterkemel k,
for 6oCoradiation. They use these, together with the kemel ke for primary kerma
to compute some simple dose distributions which were checked against published
BJR Supplement 11 data with good agreement.
When they come to implement equation (2.75) in practice, Boyer et a1 (1989)
use kemels, with these dimensions, computed (by Mackie) by Monte Carlo
methods. Additionally, for spectral beams they use five such discrete convolutions
at five different energies and add the results. The method is inherently fast
because the five convolutions are completely independent and so can be executed
in parallel if appropriate hardware is available. Boyer et a1 (1989) show how
this method can reproduce measurable depth-dose distributions for regular field
sizes to an accuracy of about 2 mm. However, the power of the method is that in
principle it can be applied to irregular fields, fields with wedges and blocks etc.
All that is required is to be able to obtain an accurate model of the fluence and the
dose follows from convolution with the total kerma kemel.
This analytic theory is undoubtedly very elegant. It carries over to radiation
dosimetry much of the formalism which is well understood in medical imaging.
However one must remember that there were some very limiting assumptions
which went into the arguments. The results are best applied to single-energy

radiation entering infinitely large uniform water phantoms. Any departure from
this must be considered with care. The analytic forms have the important merit of
giving a deep insight into the dosimetry of photon interactions. As yet, analytic
convolution methods have not been implemented into many treatment-planning
systems. It remains to be seen whether, elegant though they are, they will replace
existing routine treatment planning algorithms.
Photon dose calculation by convolution methods provide:
a a clear description of the underlying physics,
a the possibility of a very fast method of 3D dose calculation, taking advantage
of the fast Fourier transform. When the conditions are particularly simple, it is
expected that the result will have a high accuracy.
Unfortunately the theory already includes some approximations and is already
of moderate complexity when the medium is homogeneous. When similar
formalisms are attempted for dose computation in heterogeneous tissues, further
assumptions have to be made which may prove to be limitations on accuracy.
The aim is to arrive at a formalism by which the dose can be calculated without
resort to space-variant kemels, so that the advantages of calculating in Fourier
space are not lost. These advantages are very great when the dose matrix is large
(Boyer et a1 1988) and have spurred on efforts towards an analytic theory. Boyer
and Mok (1986) have provided the formalism for kerma-based convolution dose
calculations in inhomogeneous media. We shall not repeat the equations here but
give the major assumptions and limitations as specified.
Firstly, the primary fluence involves ray tracing through the heterogeneous
matrix of CT numbers, taking into account the exact fraction of each voxel
intercepted and using the voxel-dependent linear attenuation coefficient. This can
be done exactly with no loss of accuracy. The electron density of each voxel
relative to water p (r) then modifies the intensity of the electron and secondary
photon scattering in all directions in direct proportion. The primary dose D, (r)
is then a convolution of the electron-density-weighted fluence with an electron
transport kemel, but now the kemel does not take account of electron scattering
nor density-dependent attenuation. If it were intended to include these effects, the
kemel would have to be calculated by Monte Carlo methods, since Boyer and Mok
(1986) state this is too complicated to formulate analytically.
In heterogeneous media, the first scatter fluence, analogous to equation (2.67)
again has the primary fluence weighted by the relative electron density at r’, and
the exponential attenuation term becomes space variant. Boyer and Mok (1986)
show that by expanding this term in a Taylor series and neglecting high-order
terms it is possible to write down the first-scatter dose D,(r)as two convolutions
of the electron-density-weighted primary fluence with (i) the same first-scatter
kerma kemel k, (r’) as in equation (2.69) and (ii) a correction kernel, which is
I r’ I 11.’ (r’)k, (r’),acting on fraction (1 - p ( r ’ ) ) of the weighted primary
fluence. Implicit in these convolutions is (once again) the assumption that the
first-scatter kerma is untransported and set equal to the first-scatter dose.

The same assumptions are made for the multiple-scatter dose for heterogeneous
and homomogeneous tissues, namely: the photons are randomly distributed and
can be modelled with a diffusion equation; the multiple-scattercomponent is given
an average energy and this is set to be the same as for the homogeneous calculation;
the multiple-scattered kerma is not transported and is set equal to the multiple-
scatter dose. It turns out that (unlike for homogeneous tissue) the multiple-scatter
dose can not be specified in terms of a convolution with the primary fluence but
is instead a convolution of the multiple-scattered fluence with a multiple-scatter
kerma kemel. The assumption that this is spatially invariant is less worrying
because of the smaller fraction of dose which is from multiple scatter.
Boyer and Mok (1986) apply their theory to inhomogeneousphantoms with 2%
accuracy compared with measurements. They do not claim more for the theory
and point out that it is yet to be tested for irregularly shaped fields, for fields with
blocks and wedges and compensating filters.
2.5.3.1. Limitations and gains from the convolution approach. We have seen
that the three-dimensional dose distribution can be written as the convolution
integral
Dt (r) = / / / rp (r’)kt (r - r’) dV’. (2.75)
This is only possible when the kemel kt (sometimes also described as a Green’s
function) is spatially invariant. The more general form for the dose to a point r is
where E is the incident photon energy, S-2 is the incident photon direction, 2 is the
atomic number distribution in the patient and p is the density distribution in the
patient. Equation (2.77) is a superposition integral requiring the computation of a
Green’s function for every point in the patient. Clearly this is impractical since it
takes typically hundreds of VAX 780 hours to determine the Green’s function with
sufficient accuracy forjust one point. Field and Battista (1987) list the assumptions
required for replacing equation (2.77) by the more manageable equation (2.75):
1. the medium is considered of infinite extent so the kernels do not sense
boundaries;
2. the patient/phantom is considered homogeneous and water-equivalent;
3. there is no beam divergence (infinite SSD);
4. atomic number dependence is considered unimportant in the energy range
of interest;
5 . photons are monodirectional;
6. photons are monoenergetic.
Mackie has computed the Green’s function for many sets of monoenergetic
and monodirectional photons using the electron-gamma shower transport code
EGS4. The kemels were stored separately for electrons set in motion from

primary photons, first-scattered photons, second-scattered photons, multiply

scattered photons and annihilation and bremsstrahlung photons. Since the kemels
associated with these vary on different spatial scales, this method of storage can
assist in the choice of digitization used for convolution.
Indeed Mackie et a1 (1985) were possibly the first to write down expressions
whereby the dose can be computed from a convolution of fluence with what
they called dose-spread arrays. In the early work the dose-spread arrays were
separated into the dose from primary photons, the truncated first-scatter (TFS)
dose-spread array and the residual first- and multiple-scatter (RFMS) dose-spread
array. The TFS specifically excluded first-scatter dose at large distances from
the interaction site and this was incorporated into the RFMS array. Provided the
modification to the photon fluence of shielding, beam modifying devices etc, could
be modelled, the dose distribution immediately followed. Mackie et a1 (1985)
invoked O’Connor’s theorem to scale dose kemels in inhomogeneous material or
homogeneous non-water-like material (with the same atomic number Z). This
stated that the dose in two media with different densities (but the same Z) will
be the same provided all distances are scaled inversely with the density of the
material. By this means dose-spread arrays were stored in pixel steps labelled by
the product of density and length. The size of the pixel steps was different for the
different orders of scatter.
Mackie et a1 (1985) also discuss an interesting reciprocity. The dose-spread
arrays describe the transport and absorption of dose throughout a phantom due
to primary interactions in a single voxel. They also describe the absorption of
dose in a voxel due to an equal magnitude of primary interactions throughout the
volume. This leads to two different ways to compute dose from the same starting
relation depending on whether the dose is needed at just a few or many points in
the medium.
If the photon fluence is dependent only on depth into the phantom (e.g. for a
square uniform field incident normal to a plane surface), the 3D Green’s function
can be collapsed via
N N
i = l j=1
to a one-dimensional Green’s function. This reduces the convolution to a one-
dimensional convolution
Dt (z) = /r ( 2 ’ ) kt (z - z’) dz’ (2.79)
which is consequently speedier.

Boyer et ai (1988) compared dose calculation by FFT techniques with similar
computations made by ray tracing and with look-up tables. The relative speeds
depended obviously on the size of the dose matrix and on the size restrictions
applied to the convolution kemels, but to give a flavour we state:
0 ray tracing with look-up tables was some five times faster than without;

Methodsfor 2D and 3D optimization 111
a 64 x 16 x 128 dose calculation was some 80 times faster than by direct ray
tracing;
all FFT calculations obeyed the logarithmic proportionality.
On the subject of speed, Field and Battista (1987) compute the time advantage
of working in Fourier space. A one-dimensional convolution of two vectors
of length N requires N2 real multiplications. The corresponding operation in
Fourier space requires that each function be padded to length 2N. Three FFTs
are required, each needing 2N log, 2N/2 complex multiplications. 2N further
complex multiplications are needed to multiply the two transforms before back-
transformation. They assume each complex multiplication takes four times the
time for a real multiplication, so the ratio ( R I F ) , of ‘time-for-real’ to ‘time-for-
Fourier’ operation becomes
N2
(2.80)
(:), =4 (i(2N) log, (2N) + 2N)
which simplifies to
N
(2.81)
)(: = (12 log, (N) + 20) ‘
The corresponding ratios for two-dimensional and three-dimensional operations

are (deduced by squaring and then cubing N and 2N in the expression (2.80))
(2.82)
N3
(2.83)
(%>, = (14410g2(N)+176)‘
From equations (2.80X2.83) the advantage of using FFTs over real-space
convolution may be calculated. For N = 64,( R / F ) , = 0.69, ( R / F ) , = 11.6
and ( R / F ) , = 252. For N = 1024, ( R / F ) , = 7.3, ( R / F ) , = 1927 and
( R / F ) , = 664444. Clearly the advantage of using FFTs increases with the
dimensionality of the problem. (For example with N = 4 the FFT is never quicker
even in three-dimensions. In two-dimensions the break-even point exactly falls at
N = 16. In one-dimension N = 64 is below and N = 128 is above the break-even
point. In three-dimensions N = 8 is below and N = 16 is above the break-even
point.) These speed increases are well known in medical imaging.
2.5.4. Iterative 2 0 optimization by simulated annealing

Equation (2.41) shows the computation of the dose distribution D (r) from
contributions f (Y+, 4) from a number of orientations 4, In section 2.5.1 it was
shown how f (r+, 4) could be determined analytically from the density function

of point irradiations (equation (2.40)). Recently there have been successful

attempts to determine f (r$,4) by numerical inversion of equation (2.41) by
the technique of simulated annealing (Webb 1989, 1991b). The essence of this
method is that f ( r $ , # ) is built up in small increments (known as grains) by
randomly polling the orientations 4 and elements r$. Suppose that at any iteration
N the elements are such that use of equation (2.41) gives & (r) and suppose the
dose prescription is Dp (r). A quadratic cost function XN may be defined
(2.84)
The aim is to minimize XN. Hence, if addition of a grain leads to a new dose
distribution DN (r) nearer to Dp (r) than D N - ~( r ) ,then this grain is an acceptable
change. If the change AXN = XN - XN-1 is positive, the change is only accepted
with aprobability exp ( - A x N / ~ T )where
, k is the Boltzmann constant and T is a
temperature. This ensures the iterative solution does not become trapped at a local
minimum in the cost function (figure 2.30). After a very large number of iterations,
starting with high temperatures and cooling slowly, the dose distribution gradually
converges to a good approximation to the prescription. The similarity between
this process and crystal growth gives the method its name (Radcliffe and Wilson
1990). There are numerous computational arrangements which must be made for
the method to be practicable and the reader is referred to the papers for these
details. Figure 2.3 1 shows how, by combining fields from different directions with
the beam profile shaped in intensity, a high-dose volume can be made to conform
to a target with an irregular and concave border. Figure 2.32 shows a transaxial
slice with dose represented by a grey-scale. The high-dose area has been shaped
to an irregular target.
Webb (1989) applied the method to tailoring beam profiles to give dose
distributions in a plane to treatment areas with concave outlines. One advantage of
the iterative method is that the beam profiles can automatically be constrained to
be positive, leading to a practical solution. As Goitein (1990) has pointed out, this
is not an optimum solution in the sense that the dose distributionproduced does not
match the prescription exactly. It could only do that if negative beam intensities
were allowed. Since these are specifically excluded in the iterative method,
the best description of the resulting profiles is that, whilst remaining physically
realizable, they minimize the difference between the prescription and the resultant
dose distribution. Different cost functions may be used to equation (2.84) if
required and one could for example build in smoothing constraints. Morrill et al
(1991a,b) applied a similar method to optimizing the selection of wedged beams
by maximizing a cost function representing the probability of complication-free
treatment, building biological response into the calculation. Different cooling
schedules were used. The main disadvantage of the simulated annealing method
is the heavy computational burden.
We have now seen how non-uniform one-dimensional beam profiles may
be determined by either analytic and iterative deconvolution techniques or by

G
Iteration number, N
Figure 2.30. The curve shows a possible behaviour of a cost function

defined by equation (2.84)as the iterations progress. Ideally this would be
a monotonic decline to the global minimum ‘G’. However there are likely to
be local minima; two are shown at M I and M2. In order not to be trapped
in the local minima some changes must be accepted which increase the cost
function (as,for example, at ‘b’). More usually, increasing iterations decrease
the cost function (as,for example, at ‘a’). Provided some positive changes
are accepted, according to the distribution function described in the text, the
global minimum will be reached. The situation is analogous to a skier coming
downhill. The overall aim is to reach the bottom of the hill, but occasionally
the skier must ascend in order not to be trapped in local troughs.
Figure 2.31. An illustration of how multiple fields which are shaped in one
dimension can be combined to give a dose area in a slice which has an
irregular concave border. Two positions of the field are shown, labelled by
‘m’.‘k’ labels the beam element in the mrhfield at angle 0 to the anterior
direction. The dose points are labelled by (i, j ) and di,],,,,is the depth of the
( i , j ) t h dose point from the surface in the direction of the mrh beam. This
form of conformal therapy assumes the target is cylindrically symmetric (has
constant cross section).
simulated annealing. These methods may be based on dose optimization alone

or on optimizing the probability of uncomplicated tumour control. Soderstrom

Figure 2.32. Dose represented as a grey-scale. The brightest region

corresponds to the highest dose. Here the dose envelope has been shaped to
an irregular concave outline by the superposition of many one-dimensionally
modulated beams of the type shown in figure 2.31.
and Brahme (1992) have computed two quantities for each of the members of the
set of 1D profiles. The first is the entropy and the second is an integral of the low-
frequency part of the Fourier transform of each profile. They argue that a beam
with a large amount of structure will have low entropy and have a large influence
on the shape of the resulting dose distribution. Also, profiles with a large value
of the above integral in Fourier space will give a large contribution to the gross
structure of the dose distribution. Sijderstrom and Brahme (1992) thus argue that
beams with a low entropy and large value of this integral should be selected from
the set of shaped fields determining ‘preferred directions of irradiation’.
2.5.5. Iterative 3 0 optimization techniques

With modern multislice tomographic imaging techniques it is possible to create a
planning problem in three-dimensions with full 3D specification of the location of
tumour volumes and sensitive structures (see chapter 1). Treatment fields may be
shaped to the beam’s-eye-view of a target volume based on such structures using
multileaf collimators (see chapter 5). The default in this situation is to simply
weight the fields equally (Bauer 1992). However the need arises to optimize the
placement of such beams and their intensities. There is no doubt this cannot be
done by ad hoc trial and error and that computer optimization is essential. Several
workers have begun recently to tackle this problem. Mohan (1990) used the
simulated annealing technique to optimize a cost function constructed from the
dose distribution combined with tumour control probabilities and normal-tissue
complication probability data. The geometrically shaped fields were equispaced
in a coplanar arrangement together with non-coplanar positions oriented at 15’ to
30’ from the main plane of rotation. Furthermore, the possibility to modulate the

intensity of the radiation within fields was considered. No results were presented.
Simultaneously and independently, Webb (199Ob, 1991a, 1992c) constructed a
simulated annealing technique for optimizing the intensities of geometrically
shaped multileaf-collimated treatment fields with the possibility of non-coplanar
fields, A limited spatial modulation was included but the cost function did not
include the biological response. The results of applying this method to a number
of model problems were presented. It was shown that collimating the fields to
exclude the view of sensitive structures (organs at risk) ‘in line’ with the target
volume gave too poor a dose homogeneity to the target volume; an observation
also made by Mohan (1990). On the other hand, if a set of optimized weights
were determined, one per port, in achieving a uniform dose to the target volume,
the dose to sensitive tissue remained unacceptably high. Two solutions were
determined. In the first (figure 2.33), each field received two weights, one for
that part of the field seeing only target volume and the other for that part of the
field seeing both target volume and sensitive tissue. The two weights per field
were different for each BEV at each orientation, determined by the optimization.
The second solution, almost certainly impractical in the early 1990’s,was to allow
the intensity of the beam to vary at will across each BEV. Both these solutions have
been studied theoretically and gave good separation between the DVHs of target
and sensitive volumes for model problems, even when the former entwined the
latter (Webb 1992a,b,c). Bortfeld et a1 (1992), Galvin et a1 (1991), Mohan et al
(1991) and Smith et a1 (1 99 1) have also provided a solution for the inverse problem
in 3D with full modulation across the geometrically shaped beam.
The problem of determining the optimum distribution of intensity across the
multileaf collimator, given prescribed dose constraints, is formally identical to that
of determining optimum 2D radiation compensators, as tackled by Djordjevich et
a1 (1990) (see chapter 7).
At present very few treatment machines can deliver spatially modulated beams,
but one can envisage how this could be arranged by scanningtwo pairs of jaws, one
in each orthogonal direction behind the leaves of the collimator (the jaws having
a non-uniform velocity profile). This method has already been examined for
producing one-dimensional spatial modulation (Convexy and Rosenbloom 1992).
Figure 2.34 shows a possible way to deliver each field when there are two weights
per field. Mohan et a1 (1991) have very recently described how arbitrary intensity
modulation can be achieved under computer control using a multileaf collimator
and Smith et a1 (1991) also describe varying the beam intensity over a 15 x 15
matrix of field elements. Their work was applicable to multileaf collimators,
but in practice a computer-manufactured set of compensators was used. Boyer
et a1 (1991) have shown how the appropriate distribution of intensities could
be arranged by (at each orientation of the gantry) using several settings of the
multileaf collimator with the settings tracking the iso-intensity contours at the face
of the collimator (see section 2.5.6). Galvin et a1 (1991) show that for a multileaf
collimator pixellated to a 15 x 15 matrix, superposition of as few as 10 fields could
give the required intensity modulation for the problem studied. Whether this is a

beam's- eve -view
I
may be opentor closed*and w e i g h t e d
" same as"+o r "d If f eren t fro m"tt h e
region seeing T.V. only
*mode a
+mode b
imode c
tmodes b,c
Figure 2.33. Showing how to optimise the separation between the DVH for
target and sensitive volumes when multileaf-collimated BEV ports are used.
The multileaf collimator is shown to the right with the projected opening
area corresponding to the views of the target volume (TV) and the sensitive
volume (SV) (organ at risk (OAR)). These two curves overlap. The question
arises of how to handle to overlap region. Several 'modes' of irradiation are
identifred. In mode a, only the aperture corresponding to a view of the target
is irradiated. In mode b, this aperture and the overlap aperture have the same
beam-weight. In mode c, a different beam-weight is applied to the aperture
viewing target only and the overlap aperture viewing target and organ at risk.
(From Webb (1992c).)
feasible practical tool depends on how fast the fields could be reset.
The group at Galveston, Texas have used the simplex linear-programming
algorithm to optimize treatment plans (Rosen et a1 1991). This finds the best
beam weights from a set of possibilities, called the solution space, 'driven' by
constraints applied to specified points in the treatment volume. Simplex methods
become more time consuming as the number of these constraint points increases.
Recognizing that it is possibly unrealistic to specify a single dose constraint for
an organ at risk, Morrill et a1 (1991~)have introduced the binary dose- volume
constraint for organs at risk wherein the fractional volume is specified together
with the maximum dose this fraction may receive in addition to the maximum dose
which may be delivered to the rest of the organ. This defines a 'two-bin' dose-
volume histogram. Provided a large number of beams are used to aim towards
conformal therapy, it is then reasonable to assume that the high-dose volume
of any organ at risk, proximal to a target volume, will be closer to the target
volume than the low-dose region in the organ at risk. Mom11 et a1 (1991~)then
define dose-constraint points lying around collars of specified widths encircling
the target volume. The widths of the collars are determined from the dosevolume
constraints. In this way the dose-volume constraint is built into the simplex

Figure 2.34. A possible way to deliver a multileaf-collimated field when

there are two weights per field. In the lower drawing is shown the shape
of the field which encompasses the target volume ‘T’. Part of this field ‘sees’
only target volume and a smaller part ‘sees’ target plus organ at risk (‘T+O’).
Assume it is required to deliver seven units of radiation to ‘T’ and three units
of radiation to ‘T+O’.This is achieved by delivering the two separate fields
shown in the upper part of the figure. The left-hand field gives three units of
radiation to the target and organ at risk components; the right-handfield tops
up the target-only part to a total of seven units. The idea could be extended to
multiple shapedfields at the same gantry orientation, so as to create a spatial
variation of beam intensity in two dimensions (see figure 2.36).
solution. Clinical examples showed the resulting dose-volume histograms after

optimization lay entirely to the left of the two-bin prescription, indicating success.
Other methods of incorporating dose-volume constraints into linear programming
techniques have been developed by Langer and Leong (1989) and Langer et a1
( 1990).
A number of authors have developed methods of determining the beam weights
for 3D radiation therapy planning using the Cimmino algorithm (Altschuler et a1
1985, Censor et a1 1987,1988, Powlis et a1 1985, Starkshall and Eifel 1992). The
methods have generally been presented in terms of one weight per field, but there
is no inherent limitation precluding working on the general problem of fields with
spatially modulated intensities. This is not strictly an optimization so much as a
feasibility search. It works as follows. A set of points (labelled by i) spanning
the target volume, the organs at risk and other normal structures are chosen. For
those points in the target volume, a maximum and minimum dose (DMand Or)are
specified. For those points in organs at risk and other normal structure, a maximum
dose is specified. All beam-weights are constrained to be positive.
The Cimmino algorithm (Cimmino 1938) then tries to find a vector x of beam-
weights (with positive elements X j , j = 1 , 2.......N ) such that

N
Df < E A ; . , x i < DM (2.85)
i
where A ; , j is the contribution to the ith dose point from the jth beam. The
algorithm is iterative and controlled by a relaxation parameter which govems the
speed of convergence.
This method is not like simulated annealing, which is based on minimizing an
objective function based on matching dose to a prescription. The goal is not so
much to deliver a dose that matches a prescription as to find a set of conditions
which deliver a dose within the constraints. It searches for a feasible solution.
It may be there is no such feasible solution and the strength of the algorithm is
that it still provides clinically useful information in the event that the specified
dose constraints could not be satisfied by any combination of beam-weights. The
planner is presented with the result at that stage (a local minimum of the difference
between calculated doses and specified dose bounds) and can interactively make
choices whether contraints can be changed to overcome the difficulty.
A second property of the algorithm is that the importance of each dose constraint
can be weighted. In their implementation, for example, Starkshall and Eifel(l992)
weighted the dose minimum contraint 25 times more than the dose maximum, thus
signifying that it was much more important to avoid underdose than overdose to
the target volume.
Niemierko (1992) and Niemierko et a1 (1992) have developed a random search
algorithm which is able to optimize radiation dose with both physical dose and
biological endpoints and constraints. This would appear to be a particularly
powerful approach because it allows the specification to be a combination of
dose limits, desirable TCP and NTCP values and also for example dose limits
specified to fractional volumes of organs-at-risk. The specification of dose-
volume constraints is particularly interesting because some hot spots may be
allowable provided some large majority of an organ-at-risk is below a specified
limit. The approach would seem to be a very useful compromise between the
quite natural desire to specify biological endpoints (but with potentially uncertain
data and/or model) and the traditional (though surrogate) use of dose. (One might
detect here some difference of approach between the USA and the UK. In the latter
clinicians are still largely happiest specifying, and thus requesting, optimization
of dose; in the former it is becoming more common to request quantitative TCP
and NTCP for each plan.) The random search algorithm is fast and, although
not proven to converge to the global optimum, gave the same results as much
more computationally intensive optimization by simulated annealing for identical
constraints and end-points. To date this method has only been used to optimize
beam-weights and the authors’ suggestions may not be applicable to optimizing
beam apertures with a multileaf collimator.
Rosen et a1 (1991) have recently reviewed the relative merits of different
computer optimization approaches, including linear programming, the method of
simulated annealing and the feasibility technique.

2.5.6. Analytic and iterative deconvolution theory of 3 0 inverse computed

tomography
Section 2.5.1 discussed an analytic technique from Brahme (1988) to conform a
2D high-dose region to the target area. The method comprised the following steps:
1. Deliver the radiation by a full 2n rotation of a beam whose 1D intensity
variation is f (r#,$), 4 labelling the gantry orientation and rd the position
coordinate within the beam.
2. From a specified 2D dose distribution, D (r),deconvolve the rotational point-
spread dose function (PSF) d (r) to obtain the density function @ (r) of point
irradiations (equations (2.36) and (2.39)). The deconvolutioncould be performed
by an iterative method (equations (2.42) and (2.43)).
3. Forward project the density function, taking account of (spatially variable)
photon attenuation and geometry, to obtain the appropriate intensity weighting for
the connecting beam element f (rd, 4 ) (equation (2.40)).
We note that the method correctly accounted for photon attenuation in the
forward projection stage, for scattered radiation (incorporatedinto the dose kemel
(PSF)) and was constrained to give positive radiation intensities.
Boyer et a1 (1991) have elegantly extended the method to compute 3D dose
distributions by a rotational technique. The principles are identical, although the
formalism at first sight looks a little more complicated in 3D.The method again
assumes a rotational kemel A which is spatially invariant. Then the 3D dose
distribution D ( r ) is related to the density function Q by
D (r) = 111 Q (I')A (I r - I' I) d3+. (2.86)
Compare this with the corresponding 2D equation (2.36) and note the use of
symbols A and \I, instead of d and @ to distinguish the 3D quantities from their
2D counterparts (and also to follow the notation of Boyer er a1 1991). A is a
cylindrically-symmetric kernel representing dose spread in 3D.Now if D and A
are known, Q can be computed from the 3D iterative deconvolution (compare with
equations (2.42) and (2.43) in 2D):
QO(r) = a D (r) (2.87)
Qk+l ( r ) = c ( ~ k ( r ) + a k ( D ( r ) - ~ k ( r ) ~ A ( r ) ) ) (2.88)
where the subscript k refers to the kth iteration and @ means convolution. C is
a positivity constraint operator and a k controls the speed of convergence. Boyer
er a1 (1991) use some 40 iterations and calculate the convolutions using 3D FFT
techniques implemented on an array processor. This keeps the computational time
to under an hour. The convergence parameter ak assumes a subscript k indicating
that it can be varied as a function of the iteration number. At each stage of iteration

5
Figure 2.35. The coordinate system used for 3D conformal planning by

optimised rotational therapy. The beam is collimated by a multileaf collimator
whose central axis is at an angle 0 from an x-axis. The ray from the source
<
at s to the location r in the patient lies at angles and 77 as shown. The
intensity function where this ray intersects the collimator is specified by
f (< ( O ) , q ( 8 ) ) . (From Boyer eta1 (1991).)
the difference A Dk (r) between the dose prescription D (r) and the kth estimate
Dk ( r ) of the dose distribution is
Minimizing
Xi+l = (ADk+l (r))' (2.90)
r
by setting to zero the derivative of this function with respect to ak gives
The collimator for the radiation is specified by its angle 0 from an x-axis;
the ray from the source directed at vector r has angle f and q to the central
ray directed towards the origin, through which the axis of rotation passes (figure
2.35). The intensity variation at the collimator is then specified by the function
f (t (0) , q (0)) which is given by ray tracing through the attenuating medium to
the source position:
where s (e)represents the position of the source when the gantry is at angle 6' and
labels the direction of the ray, along which the line integral is taken.
a!

Figure 2.36. Showing how it might be possible to use multiple fields at

any given orientation, each shaped by a multileaf collimator and tailored to
conform to the iso-intensity contours across the face of the openfield. In this
schematic it is implied that the smallest inner part of the field receives ' I O
units' of radiation intensity, the annulus outside this receives '8 units' and so
on. A few leaves are drawn to give an idea of how the multileaf collimator is
used.
Boyer et a1 (199 1) thus develop the formalism for computing a spatially varying
intensity across the face of the multileaf collimator. They argue that in practice
these irradiations could be built up at each gantry orientation as a sequence of
small exposures with different multileaf collimator settings defined by the shape of
the iso-intensity contours at the plane of the collimator (figure 2.36). This would
require a computer-controlled multileaf collimator. At present it is not feasible
to implement the method because of the very large number of orientations the
multileaf collimator must take up, and also the large number of sub-fields needed
at each orientation. Indeed Boyer et a1 (1991) consider this may not even be
developed in the near future. Boyer et a1 (1991) use very small (1") angular
increments for the gantry to study the limiting performance of such a method. This
is valuable work which helps to establish just how much effort is worth putting
into rotational conformal therapy with a multileaf collimator.
They studied several cylindrically-symmetric distributions and came to the
somewhat disappointing conclusion that the tails of the skirts of the point-dose
kemel led to rather large doses being delivered to normal tissue if this were
enclosed by target volume. However, the method coped well with delivering a
conformal dose to a volume which had a concave outline. Boyer et a1 (1991)
discuss the possibility of using non-coplanar multileaf-collimator-defined fields
with spatially varying beam intensities, removing the cylindrical symmetry of the
dose kemel A. This might lead to better therapeutic ratios. The same conclusion
was reached by Webb (1992a,b,c), who computed the spatially varying intensities
by a numerical optimization method for non-coplanar fields with a simple dose
calculating algorithm.
2.5.7. The inversion algorithm for proton therapy

So far this chapter has been concerned with optimizing photon therapy. High-
energy protons may also be used for precision conformal therapy. The subject

of proton therapy will be discussed in detail in chapter 4. Here we briefly

present a theory of optimization which has formal similarities with analytic photon
optimization. Figure 2.37 shows a beam of protons emanating from a cyclotron
and entering the patient, depositing energy along their track towards a Bragg peak.
Suppose the aim is to produce a high-dose volume in the re-entrant volume, shown
with a solid line. This can be achieved if the proton beam can make a linear
traverse from one side of the region to the other, and at the same time the Bragg
peak may be adjusted to different depths z at each position of the beam during
its linear traverse. (Alternatively, if the proton energy is fixed (Bragg peaks all
at the same depth) to treat the volume requires rotation as well as translation, as
for photons). Consider the case with a single, translated, beam portal. Suppose
the dose distribution from a narrowly collimated monoenergetic proton beam is
h (r) and the desired dose distribution is D (r). If the impulse response h may be
considered spatially invariant, then, as for photons, we may write
D (r) = 11 1 CD (r’)h (r - r’)d3r’. (2.93)
i.e. the dose distribution is a convolution of the elementary distribution h with

the density @ of such elementary distributions. (Compare with equation (5.6)
Figure 2.37. Arrangement for proton therapy with scanned elementary

beam. The soiid butterfly shaped curve is the treatment area. The elementary
beam range is modulated by double-wedge filters. The beam is swept through
the target. (From Brahme et a1 1989).)

Figure 2.38. Isometric plot of ( a ) the elementary proton beam, (b) the
desired dose in the target area, and (c) the density of Bragg peaks. (From
Brahme et a1 ( 1 989/.)
(chapter 5 ) for single-port photon therapy with a multileaf collimator.) The

irradiation density function may be obtained in exactly the same way as for
photons by equations (2.42) and (2.43), where again positivity is easy to invoke
(see also equations (5.7) and (5.8) of chapter 5).
Brahme et a1 (1989) introduced this method, applying it to a number of model
problems. For example figure 2.38 shows the elementary beam profile, the dose
prescription and the density function. In this figure the beam enters top right. As
one expects, the density function peaks at the distal side of the target volume. It
is easy to see why. Proton beams with Bragg peaks in this far-distal region will
deposit some energy along the track back to the source. Hence the density of
elementary beams with Bragg peaks nearer to the source will be lower because
these regions already pick up some radiation from the distal-side Bragg peaks.
The disadvantage of this single-port technique is that tissue outside the treatment

volume, but through which the radiation passes, will obviously be irradiated. This
can be overcome by using rotation methods or multiple ports, as discussed by
Brahme et a1 (1989).
The function @ (r) gives the density of Bragg peaks inside the target volume.
At each translation position of the beam, the range must be modified according
to this distribution. The range modulation may be achieved by using beam filters
(shown as wedges of lead and beryllium in figure 2.37). At every lateral position
of the elementary beam port, a set of differentfiltrations are required with the beam
monitor units given by the appropriate irradiation density for each of the points
in- line down that elementary track. In practice this would be quite complicated
and it may be simpler to keep the range constant and use rotation instead.
The heart of the technique is the elementary function h. This is given by
(2.94)
where r is the radial position, z is distance along the central axis of the beam, D ( z )
is the central axis depth dose of a broad beam and (r2( z ) ) is the mean square radius
at depth z (see also chapter 4).
2.6. SUMMARY
We began this chapter with a consideration of aspects of two-dimensional

radiotherapy planning and investigated the intensity modulation needed for beams
to combine to give a high-dose region which might not have a convex boundary.
This enabled a deeper insight into what had been achieved in the past with
unmodulated but partially blocked fields in gravity-assistedrotation therapy. The
concept of the point-dose kernel naturally arose and is at the heart of a number of
new developments, generically called convolutiondosimetry. Whilst these are not
yet widely established in routine treatment planning, they are excellent benchmark
tools and are expected to play a wider role as computers become ever faster. There
have been several attempts to determine how multiple 2D fields, shaped by a
multileaf collimator and not necessarily coplanar, might be combined to give a
highly conformal 3D dose distribution. Some of these are analytic approaches and
others are iterative. The parallel approach in proton therapy was reviewed here,
ahead of the main discussion of this topic, because of its formal similarity with
optimizing photon therapy.
In the next chapter conformal stereotactic radiotherapy will be discussed and
the following chapter will retum to discussing the physics of radiotherapy with
protons.

Appendix 2A. A historical note on the origins of rotation therapy 125
2.7. APPENDIX 2A. A HISTORICAL NOTE ON THE ORIGINS OF

ROTATION THERAPY
In this chapter we have seen that at the heart of conformal therapy lies the need to
arrange for multiple fields at different orientations to the patient. Some methods
can be formalized in terms of a rotation model. Rotation therapy has a long history
going back to the germ of an ideanot long after the turn of the century (Baese 1915,
Portes and Chausse 1921). More serious attempts were made in the 1930’s and
rotation therapy was well established by the 1940’s and had become very elaborate
by the early 1950’s.
To understand the popularity of what is now a rarely used technique, one has to
remember that high-energy therapy implied in those days the use of photons at or
below some 2 MV, i.e. the radiation was considerably less penetrating than that in
common use today from linear accelerators. Often it was as low as 250 kV. The
aim was to increase the ratio between the dose to the target volume and the dose to
healthy normal tissue through which the radiation had to pass, by rotating a beam
about an axis centred in the target volume.
We should also remember that techniques for establishing the target volume
were much poorer than today (no CT or MRI), that computers for dose planning
were unheard of, that treatment times per fraction could be large and that
techniques for positioning the patient were not particularly accurate. Today
they would probably be considered quite unacceptable. Against these difficulties
rotation therapy persevered and the technique became widely established, building
up a huge literature.
Consider for example dose planning. Hand calculations were tedious enough
for combining just a few beams. Rotation therapy presented a considerable
problem. The source-to-skin distance was continuously varying. Establishing
the patient contour was itself difficult (Friedman et a1 1955). Generally, tissue
inhomogeneities were ignored (Wheatley 1955). As a result there was much resort
to experimental dosimetric verification with phantoms. For example films were
placed in cylinders of scattering material (Lantzl and Skaggs 1955, Friedman
et a1 1955, Peterson et a1 1955) or ion-chamber measurements were made
(Braestrup and Mooney 1955). These gave a crude indication of the expected dose
distributions, but, since the phantoms were not customized to match the individual
external patient contour and they were homogeneous, the results were wrong.
Transit-dose measurements were used to assess by how much they were wrong.
A dosemeter was placed well downstream of the patient to record the primary
exit radiation only at each orientation of the patient (Pfalzner 1956) and these
measurements, together with knowledge of the unattenuated beam intensity, were
used to deduce the ‘equivalent water thicknesses’ at each angle. This naturally led
to plots of the ‘equivalent contour’ for the patient. Fedoruk and Johns (1957) built
a special transit-dose collimator to exclude scattered radiation when making such
measurements. In 1953 Smithers at the (then) Royal Cancer Hospital accepted the
offer of help from a professional sword swallower (figure 2.39). A dosemeter was

thus placed in the thorax of a living subject rotating in an x-ray field to assess
the limitations of his models for dosimetry (Wheatley et a1 1953). However,
imaginative though this was as an example of early in-vivo dosimetry, it was
hardly a practical possibility for all subjects! Accurate dosimetry remained a real
problem.
Localization of the target volume was generally estimated from planar
radiographs, classical axial transverse tomography or fluoroscopy. Some
manufacturers produced diagnostic equipment which mimicked the treatment
Figure 2.39. Showing how Smithers used the services of a professional

sword swallower in the 1950s to try to estimate dosimetry in the thorax during
rotation therapy. (From Smithers’ papers.)

References 127
geometry and shared a room with it (see figure 6.3) (Klemm 1957). After target
localization, the patient was wheeled across on the couch for treatment. This was
appropriate for rotation therapy with the patient in the horizontal position (SO-
called paraxial conical therapy).
We might be more concemed at the common practice of rotating the patient
vertically in a stationary beam. Possibly this was influenced by the then common
practice of performing axial transverse tomography in the vertical position, which
itself became popular from the need to image fluid levels in the lungs of patients
with tuberculosis. Imagine the difficulty of ensuring the target volume was
properly located in the beam, of reproducing the patient positioning accurately
and of asking the patient to sit quietly for some 10-15 minutes during irradiation
(Pfalzner and Inch 1956, Berman, 1957). Errors must have occurred. The
difficulties were well known to the workers of the time but they could do little
about them. There was much disagreement in the literature over the acceptability
of so called circumaxial (i.e. vertical) rotation therapy.
Rotation therapy with a limited arc of travel was sometimes known as pendulum
therapy, there being provided a means of reversing the direction of motion of
the x-ray source at two end-stops defining the arc of travel (Busse 1956). A
form of pendulum therapy was devised by Lupo and Pisani (1956) with apparatus
which was the therapy equivalent of classical planigraphy. Some considered the
problems would be better solved with the patient recumbant and the necessary
rotation arcs obtained by combinations of translations and pivoting of the x-ray
source (Peterson et a1 1955).
For all this the intentions were fine and indeed a quotation from a paper by
Friedman et a1 (1955) is still echoed today in introducing the goals of conformal
therapy:
‘The efficiency of a particular form of radiation therapy is gauged by its ability
to produce rather homogeneous dose distribution within the tumour region with
minimal irradiation of the surrounding tissues.’
REFERENCES
Ahnesjo A 1987 Invariance of convolution kemels applied to dose calculations

for photon beams The use of computers in radiation therapy ed I A D Bruinvis
- 1989 Collapsed cone convolution of radiant energy for photon dose
calculation in heterogeneous media Med. Phys. 16 577-592
- 1990Photon beam characterizationfor dose calculations from first principles
The use of computers in radiation therapy: Proceedings of the 10th Conference
ed S Hukku and P S Iyer (Lucknow) pp 64-67
Ahnesjo A, Andreo P and Brahme A 1987a Calculation and application of point
spread functions for treatment planning with high energy photon beams Acta
Oncol. 26 49-56

Ahnesja A and Mackie R 1987b Analytical description of Monte Carlo generated

photon dose convolution kemels The use of computers in radiation therapy ed
I A D Bruinvis et a1 (Roc. 9th ICCRT) pp 197-200
Ahnesj0 A and Trepp A 1991 Acquisition of the effective lateral energy h e n c e
distribution for photon beam dose calculations by convolution methods Phys.
Med. Biol. 36 973-985
Altschuler M D, Powlis W D and Censor Y 1985 Teletherapy treatment planning
with physician requirements included in the calculation: 1. Concepts and
methodology Optimisation of cancer radiotherapy ed B R Paliwal, D E Herbert
and C G Orton (New York: AIP (AAPM)) pp 443452
Baese C 1915 Method of, and apparatus for, the localisation of foreign objects in,
and the radiotherapeutic treatment of, the human body by the x-rays UK Patent
no 100491
Barth N H 1990 An inverse problem in radiation therapy Znt. J.Rad. Oncol. B i d .
Phys. 18 425-43 1
Bauer J, Hodapp N and Frommhold H 1992 A comparison of dose distributions
with and without multileaf-collimator for the treatment of oesophageal
carcinoma Proc. ART91 (Munich) (abstract book p 49) Advanced Radiation
Therapy: Tumour Response Monitoring and Treatment Planning ed A Breit
(Berlin: Springer) pp 649-653
Berman H L 1957 Immobilisation of the head during rotational x-ray therapy
Radiology 68 579-581
Bortfeld T, Boesecke R, Schlegel W and Bohsung J 1990b 3D dose calculation
using 2D convolutions and ray tracing methods The use of computers in
Iyer (Lucknow) pp 238-24 1
Bortfeld Th, Burkelbach J, Boesecke R and Schlegel W 1990a Methods of image
reconstruction from projections applied to conformation therapy Phys. Med.
Biol. 35 1423-1434
Bortfeld T, Burkelbach J and Schlegel W 1992 Solution of the inverse problem in
3D Proc. ART91 (Munich)(abstracts book p 46) Advanced Radiation Therapy:
Tumour Response Monitoring and Treatment Planning ed A Breit (Berlin:
Springer) pp 503-508
Boyer A L, Desobry G E and Wells N H 1991 Potential and limitations of invariant
kernel conformal therapy Med. Phys. 18 (4) 703-7 12
Boyer A and Mok E 1985 A photon dose distribution model employing
convolution calculations Med. Phys. 12 169-177
-1986 Calculation of photon dose distributions in an inhomogeneous medium
using convolutions Med. Phys. 13 503-509
Boyer A L, Wackwitz R and Mok E C 1988 A comparison of the speeds of three
convolution algorithms Med. Phys. 15 224-227
Boyer A L, Zhu Y, Wang L and Francois P 1989 Fast Fourier transform
convolution calculations of x-ray isodose distributions in homogeneous media
Med. Phys. 16 248-253

References 129
Brace J A 1985 Computer systems for the control of teletherapy units Progress in
medical radiation physics:2 ed C G Orton (New York: Plenum) pp 95-1 11
Braestrup C B and Mooney R T 1955 Physical aspects of rotating telecobalt
equipment Radiology 64 17-27
Brahme A 1987 Optimisation of conformation and general moving beam radiation
therapy techniques The use ofcomputers in radiation therapy ed I A D Bruinvis
- 1988 Optimisation of stationary and moving beam radiation therapy
techniques Radiother. Oncol. 12 129-140
- 1992 Recent developments in radiation therapy planning Proc. ART91
(Munich)(abstracts book p 6) Advanced Radiation Therapy: Tumour Response
Brahme A, Kallman P and Lind B K 1989 Optimisation of proton and heavy ion
therapy using an adaptive inversion algorithm Radiother. Oncol. 15 189-197
Brahme A, Roos J E and Lax I 1982 Solution of an integral equation encountered
in rotation therapy Phys. Med. Biol. 27 1221-1229
Busse W 1956 Einstellung und Dosisbestimmung bei der Pendelbestrahlung der
Parametrien Strahlentherapie 101 400-404
Casebow M 1990a A modified synchronous shielding technique in rotation
therapy Brit. J . Radiol. 63 482-7
-1990b private communication.
Censor Y, Altschuler M D and Powlis W D 1988 On the use of Cimmino’s
simultaneous projections method for computing a solution of the inverse
problem in radiation therapy treatment planning Inv. Prob. 4 607-623
Censor Y, Powlis W D and Altschuler M D 1987 On the fully discretised model
for the inverse problem of radiation therapy treatment planning Proc. 13th
Ann. Northeast Bioengineering Conference ed K R Foster (IEEE 87-CH2436-4
Library of Congress catalog card number 87-80502) pp 211-214
Cimmino G 1938 Calcolo approssimato per le soluzioni die sistemi di equazioni
lineari La Ricerca Scientifica Roma 1938-XVI,Ser 11, Anno IX, Vol I pp 3 2 6
333
Convery D J and Rosenbloom M E 1992 The generation of intensity-modulated
fields for conformal radiotherapy by dynamic collimation Phys. Med. Biol. 37
1359-1374
Cormack A M 1987 A problem in rotation therapy with X-rays Znt. J . Rad. Oncol.
B i d . Phys. 13 623-630
-A M 1990 Response to Goitein Znt. J . Rad. Oncol. B i d . Phys. 18 709-710
Cormack A M and Cormack R A 1987 A problem in rotation therapy with X-rays
2: dose distributions with an axis of symmetry Znt. J . Rad. Oncol. Biol. Phys.
13 1921-1925
Cormack A M and Quinto E T 1989 On a problem in radiotherapy: questions of
non-negativity Int. J . Imaging Systems and Technology 1 120-124
Cunningham J R 1982 Tissue inhomogeneity corrections in photon-beam
treatment planning Progress in medical radiation physics.1 ed C G Orton (New

York: Plenum) pp 103-1 3 1

Davy T J 1985 Physical aspects of conformation therapy using computer-
controlled tracking units Progress in medical radiation physics:2 ed C G Orton
(New York: Plenum) pp 45-94
Djordjevich A, Bonham D J, Hussein E M A, Andrew J W and Hale M E 1990
Optimal design of radiation compensators Med. Phys. 17 (3) 397-404
Fedoruk S 0 and Johns H E 1957 Transmission dose measurement for cobalt 60
radiation with special reference to rotation therapy Brit. J . Radiol. 30 190-195
Field C and Battista J J 1987 Photon dose calculations using convolution in real
and Fourier space: assumptions and time estimates The use of computers in
radiation therapy ed I A D Bruinvis et a1 (Proc. 9th ICCRT) pp 103-106
Friedman M, Hine G J and Dresner J 1955 Principles of supervoltage (2 million
volts) rotation therapy Radiology 64 1-16
Galvin J M, Smith R M and Chen X-G 1991 Modulation of photon beam intensity
with a multileaf collimator Proc. 1st biennial ESTRO meeting on physics in
clinical radiotherapy (Budapest, 1991) p 5
Goitein 1990 The inverse problem Int. J . Rad. Oncol. Biol. Phys. 18 489491
Green A 1959 A technical advance in irradiation technique Proc. R. Soc. Med. 52
344-346
-1965 Tracking cobalt project Nature 207 1311
Herman G T 1980 Image reconstruction from projections: The Fundamentals of
Computed Tomography (New York: Academic)
Ilfeld D N, Wright K A and Salzman F A 1971 Synchronous shielding and
field shaping for megavolt irradiation of advanced cervical carcinoma Am. J .
Rontgenol. 112 792-796
Jennings W A 1985 The tracking cobalt project: from moving beam therapy
to three dimensional programmed irradiation Progress in medical radiation
physics:2 ed C G Orton (New York: Plenum) pp 1-44
Jones D 1990 Light engineering Nature 343 5 16
Kallman P, Lind B K and Brahme A 1992 An algorithm for maximising the
probability of complication-free tumour control in radiation therapy Phys. Med.
Biol. 37 87 1-890
Klemm T 1957 Localisation and setting techniques in moving-field therapy SRW
News: J . Siemens-Reiniger-Werke AG, Erlangen 1 8-12
Kom G A and Kom T M 1968Mathematical handbook for scientists and engineers
(New York: McGraw Hill) p 564
Lakshminarayanan A V 1975 Reconstruction from divergent beam data Technical
Report No 92 (State University of New York at Buffalo, Department of
Computer Science)
Langer M, Brown R, Urie M, Leong J, Stracher M and Shapiro J 1990 Large
scale optimisation of beam weights under dose-volume restrictions Int. J . Rad.
Oncnl. Biol. Phys. 18 887-893
Langer M and Leong J 1989 Optimisation of beam weights under dose-volume
restrictions Int. J . Rad. Oncol. Biol. Phys. 13 1255-1259

References 13 1
Lantzl L H and Skaggs L S 1955 Revolving beam isodose distributions from a

cobalt 60 therapy unit Semiannual report to the Atomic Energy Commission,
Argonne
Lax I and Brahme A 1982 Rotation therapy using a novel high-gradient filter
Lind B K 1990 Solutions of inverse problems in radiation therapy The use of
computers in radiation therapy: Proceedings of the 10th Conference ed S
Hukku and P S Iyer (Lucknow) pp 163-166
Lind B K and Kallman P 1990 Experimental verification of an algorithm for
inverse radiation therapy planning Radiother. Oncol. 17 359-368
Lind B K and Brahme A 1992 Photon field quantities and units for kemel based
radiation therapy planning and treatment optimisation Phys. Med. Biol. 37 89 1-
909
Lupo M and Pisani G 1956 Une variante technique de la Stratitherapie pseudo-
pendulaire a la palmieri J . Radiol. et &Elect. 37 39-41
Mackie T R, Ahnesjo A, Dickof P and Snider A 1987 Development of a
convolution / superposition method for photon beams The use of computers
in radiation therapy ed 1.A D Bruinvis et a1 (Proc. 9th ICCRT) pp 107-1 10
Mackie T R, Reckwerdt P J, Gehring M A, Holmes T W, Kubsad S S , Sanders CA,
Paliwal B R and Kinsella T J 1991 A fast convolution/ superposition algorithm
for radiation therapy treatment planning Proc. 1st biennial ESTRO meeting on
Mackie T R, Reckwerdt P J, Gehring M A, Holmes T W, Kubsad S S,Thomadsen
B R, Sanders C A, Paliwal B R and Kinsella T J 1990a Clinical implementation
of the convolution/superposition method The use of computers in radiation
therapy: Proceedings of the 10th Conference ed S Hukku and P S Iyer
(Lucknow) pp 322-325
Mackie T R, Reckwerdt P J, Holmes T W and Kubsad S S 1990b Review of
convolution/superpositionmethods for photon beam dose computation The use
of computers in radiation therapy: Proceedings of the 10th Conference ed S
Hukku and P S Iyer (Lucknow) pp 20-23
Mackie T R, Scrimger J W and Battista JJ 1985 A convolution method for
calculating dose for 15-MV x rays Med. Phys. 12 (2) 188-196
Mohan R 1990 Clinically relevant optimisation of 3D conformal treatments The
S Hukku and P S Iyer (Lucknow) pp 36-39
Mohan R, Chui C and Lidofsky L 1986 Differential pencil beam dose computation
model for photons Med. Phys. 13 6 4 7 3
Mohan R, Mageras G and Podmaniczky K C 1991 A model for computer-
controlled delivery of 3D conformal treatments Med. Phys. 18 612
Mom11 S M, Lane R G, Jacobson G and Rosen I 1991a Treatment planning
optimization using constrained simulated annealing Phys. Med. Biol. 36 1341-
1361
Mom11 S M, Lane R G and Rosen I 1991b Constrained simulated annealing for

optimised radiation therapy treatment planning Comp. Methods Prog. Biomed.

33 135-144
Morrill S M, Lane R G, Wong J A and Rosen I I 1991c Dose-volume
considerations with linear programming optimisation Med. Phys. 18 1201-
1210
Niemierko A 1992 Random search algorithm (RONSC) for optimisation of
radiation therapy with both physical and biological endpoints and constraints
Niemierko A, Urie M and Goitein M 1992 Optimisation of 3D radiation therapy
with both physical and biological endpoints and constraints Int. J . Rad. Oncol.
Biol. Phys. 23 99-108
Nilsson M and Knoos T 1991 Application of the Fano theorem in inhomogeneous
media using a convolution algorithm Proc. 1st biennial ESTRO meeting on
Peterson 0 S, Foley J C and Mosher R F 1955 Device for control of radiation
distribution in moving beam therapy Radiology 64 412416
Pfalzner P M 1955 Rotation therapy with a cobalt 60 unit: 2: Transit dose
measurements as a means of correcting dose for non-waterequivalent absorbing
media Acta Radiologica 45 62-68
Pfalzner P M and Inch W R 1956 Rotation therapy with a cobalt 60 unit: 1:Physical
aspects of circumaxial rotation Acta Radiologica 45 52-61
Portes F and Chausse M 1921 ProcCdt pour la mise au point radiologique sur un
plan sCcant d’un solide, ainsi que pour la concentration sur une zone dtterminte
d’une action radiothtrapeutique maximum, et dispositifs en permettant la rC
alisation French Patent 54 1914
Powlis W D, Altschuler M D and Censor Y 1985 Teletherapy treatment
planning with physician requirements included in the calculation: 2. Clinical
Applications Optimisation of cancer radiotherapy ed B R Paliwal, D E Herbert
and C G Orton (New York AIP (AAPM)) pp 453-461
Proimos B S 1960 Synchronous field shaping in rotational megavolt therapy
-1961 Synchronous protection and field shaping in cyclotherapy Radiology
77 591-599
-1963 New accessories for precise teletherapy with 6oCounits Radiology 8 1
307-3 16
-1969 Shaping the dose distribution through a tumour model Radiology 92
130-1 35
Proimos B S, Tsialas S P and Coutroubas S C 1966 Gravity oriented filters in arc
cobalt therapy Radiology 87 933-937
Radcliffe N and Wilson G 1990 Natural solutions give their best New Scientist
( 14thApril 1990) pp 47-50
Rawlinson J A and Cunningham J R 1972 An examination of synchronous
shielding in “CO rotational therapy Radiology 102 6 6 7 6 7 1
Rosen I I, Lane R G, Mom11 S M and Belli J A 1991 Treatment plan optimisation

References 133
using linear programming Med. Phys. 18 141-152

Smith R M, Galvin J M and Chen X G 1991 The use of a multileaf collimator to
optimise dose distributions Med. Phys. 18 613
Smithers D 1953 Collected papers and correspondence of Sir David Smithers held
in the library archives of the Institute of Cancer Research, Sutton
Soderstrom S and Brahme A 1992 Selection of suitable beam orientations in
radiation therapy using entropy and Fourier transform measures Phys. Med.
Biol. 37 91 1-924
Starkschall G and Eifel P J 1992 An interactive beam-weight optimisation tool for
three-dimensional radiotherapy treatment planning Med. Phys. 19 155-163
Takahashi S, Kitabatake T, Morita K, Okajima S and Iida H 1961 Methoden zur
besseren Anpassung der Dosisverteilung an tiefliegende Krankenheitsherde bei
der Bewegungsbestrahlung Strahlentherapie 115 478-488
Trump J G, Wright K A, Smedal M I and Salzman F A 1961 Synchronous field
shaping and protection in 2-million volt rotational therapy Radiology 76 275
Webb S 1988 The mathematics of image formation and image processing The
Physics of Medical Imaging ed S Webb (Bristol: Adam Hilger) p 560
- 1989 Optimisation of conformal radiotherapy dose distributions by
simulated annealing Phys. Med. Biol. 34 1349-1369
-1990a Inverse tomograph Nature 344 284
-1990b A new dose optimising technique using simulated annealing P roc. 9th
Ann. Meeting of ESTRO (Montecatini,1990) p 263
- 199Oc From the watching of shadows: the origins of radiological
tomography (Bristol: Adam Hilger) p 125 et seq
-199l a Optimisation by simulated annealing of three-dimensional conformal
treatment planning for radiation fields defined by a multileaf collimator P hys.
Med. Biol. 36 1201-1226
- 199l b Optimisation of conformal radiotherapy dose distributions by
simulated annealing 2: inclusion of scatter in the 2D technique Phys. Med. B i d .
36 1227-1237
-1992a Optimisation by simulated annealing of three-dimensional conformal
treatment planning for radiation fields defined by a multileaf collimator: 2.
Inclusion of two-dimensional modulation of the X-ray intensity Phys. Med.
Biol. 37 1689-1704
- 1992b Optimising dose with a multileaf collimator for conformal
radiotherapy Proc. 50th Ann. Congress of the British Institute of Radiology
(Birmingham,1992) (London: BIR)p 15
- 1992c Optimised three dimensional treatment planning for volumes with
concave outlines, using a multileaf collimator P roc. ART91 (Munich)(abstract
book p 66) Advanced Radiation Therapy: Tumour Response Monitoring and
Treatment Planning ed A Breit (Berlin: Springer) pp 495-502
Wheatley B M 1955 A method of dose calculation with applications to moving-
field therapy Brit. J . Radiol. 28 566-573
Wheatley B M, Steed P R, Savage E W, King J H, Forster E W, Hodt H J, Jones

I R and Smithers D W 1953 The two million volt Van de Graaf generator
installation for rotation therapy at the Royal Cancer Hospital Brit. J . Radiol.
26 57-72
Zhu Y and Boyer A 1990 X-ray dose computations in heterogeneous media using
3-dimensional FFT convolution Phys. Med. Biol. 35 351-368
Zhu Y, Boyer A L and Desobry G E 1992 Dose distributions of x-ray fields as
shaped with multileaf collimators Phys. Med. Biol. 37 163-173

CHAPTER 3
STEREOTACTIC RADIOSURGERY AND

RADIOTHERAPY
3.1. INTRODUCTION
Tumours of the brain are regarded as special candidates for precision conformal
therapy. Indeed any interventional procedure, be it a biopsy, surgery, the
implantation of radioactive material or extemal beam therapy must be carried out
as accurately as possible. Stereotactic procedures have developed to meet these
needs. The term arises from stereo (the Greek word for three-dimensional) and
either taxic (the Greek word for system or arrangement), or the Latin verb tactus
(meaning to touch) (Galloway and Maciunas 1990). The special need for precision
arises because the brain is unique in being almost entirely surrounded by bone,
with the associated requirement for the surgeon to preserve this protection and the
surgical constraint of preserving the function of tissues surrounding the surgical
site. For these reasons using radiation to achieve the same aims as conventional
surgery is attractive (section 3.2).
Three-dimensional tomographic imaging is specially essential to these activi-
ties. Indeed neurological procedures have been transformed by the availablility of
a multiplicity of imaging modalities such as CT,MRI and PET.
It follows that accurate registration of image data from different modalities
and between these and radiotherapy and surgical equipment is a sine qua
non. Whilst it is sometimes possible to achieve this with landmark-based
correlations, computing rigid and/or elastic transformations(see chapter l), a class
of procedures has developed based on the use of the stereotactic frame which
‘locks’ the head into the same position on imaging and treatment machines (figure
3.1).
3.1-1. Stereotactic positioning devices

It would appear that the first stereotactic frame was designed for investigating
animal brains by the British surgeon Robert Henry Clarke and built by James
Swift in London in 1905 (Horsley and Clarke 1908). The apparatus was
135
136 Stereotactic radiosurgery and radiotherapy
Figure 3.1. The principle of stereotactic irradiation. The patient wears a

frame ( F ) which may be bolted to the calvarium and is attached to the patient
support system (Pi. The frame is worn both for three-dimensional medical
imaging (C could be an x-ray CT,MRI, SPECT or PET scanner) and, later, at
the time of treatment (T could be a treatment linac or a gamma knife). The
important function of the Z-bars ( Z ) is described in the text.
patented by Clarke in 1914 (Clarke 1914) and can still be seen in the museum
of University College Hospital (Fodstad et a1 1990). Not until 1947 was
apparatus of this kind used for stereotactic intracranial surgery in man (Spiegel
et a1 1947). Leksell (1949) revitalized interest by designing a new stereotactic
frame for precision cerebral surgery which formed the basis for his innovative
proposal for conducting radiosurgery (see section 3.2) shortly afterwards (Leksell
1951). Modem developments in frame construction stem from this initiative.
These are numerous and to avoid tedium only some representative examples are
reviewed. In particular we may note that the well known stereotactic frames
were developed for conventional surgery, biopsy, the placement of electrodes
etc, and whilst some workers have used them for radiosurgery there has been a
tendency to develop special frames for this purpose. So, important and numerous
though they are, we simply note in passing the names of the four best known
of these frames-Leksell apparatus, Riechert-Mundinger apparatus, Todd-Wells
apparatus, Brown-Robens-Wells (BRW) apparatus. These are described in detail
by Galloway and Maciunas (1990). Instead we turn to some of the special-purpose
constructions. Not all these have been used for stereotactic radiotherapy-
some were just for image registration-but it is useful to group them here when
considering stereotactic procedures in general.
The principles of stereotactic image registration are well illustrated by
Bergstrom et a1 (1 98 1). Their stereotactic frame comprised an aluminium base-
plate with various knobs and holes which during scanning was bolted to a special
head-holder on the scanner couch. Four aluminium bars were attached to the base-
plate in an adjustable way so they could be positioned as close to the head as

Introduction 137
Figure 3.2. Showing the patient head immobilization with base-plate,

aluminium bars and glass-fibre mould, as developed by Bergstrom. (From
Bergstrom et a1 (1981 ).)
possible. Either small aluminium screws then attached these bars to the calvarium
or the head was immobilized by a glass-fibre mesh ('Lightcast'), hardened by
ultraviolet light, which wrapped around the aluminium bars (figure 3.2), and by
incorporating a dental bite (Kingsley et a1 1980). Wrapping the head is not a
disadvantage for radiosurgery, unlike for stereotactic surgery where the method
of fixing with screws is preferred, leaving a wide area of the skull available for
access.
A special localization frame made of perspex and aluminium was then attached
to the base-plate (figure 3.3). This frame contained thin aluminium strips running
at 45" and 90" to the baseplate. Running parallel to these strips were thin plastic
tubes into which radioactive fluid could be placed. This arrangement is sometimes
referred to as a 'Z' bar or 'N bar' because the positions of the strips and tubes have
the shape of these letters.
The patient wore the device during scanning (figure 3.4). The reproducibility
of patient positioning was better than 3 mm (Kingsley et a1 1980). Both the CT
scanner and the PET scanner had equivalent head-holders to which the base-plate
could be attached. The material of the head holder and the localization frame
did not produce image artefacts. The thin aluminium strips showed up in the CT
images and the images of the radioactive fluid in the plastic tubes showed up in
the PET images. In this way the position of each transaxial slice was coded into
each image.
Stereotactic coordinates were established as follows: The vertex of the 'N' bar
defined the y = 0 plane. This was a fixed distance above the baseplate. The
distances between the (point) images of the two strips at 45" to each other in any
transaxial slice gave the y coordinate of this transaxial slice. More generally, for
an arrangement where the angle between horizontal and diagonal bars is not 45", if
I 1 is the distance between the images of the two strips which meet at the baseplate
and 12 is the distance between the images of the other two strips meeting more

I 38 Stereotactic radiosurgery and radiotherapy
Figure 3.3. A localization frame made from perspex with embedded

aluminium strips. The two strips at 45" are used to find the slice level ( y
coordinate). The corner of the angle is the zero reference plane. A thin plastic
tube is attached to theframe to take a radioactivefluid for when the system is
used during PET scanning. It could also take a fluid such as copper sulphate,
if theframe were used during magnetic resonance imaging. (FromBergstrom
et a1 ( I 9 8 1 ) ~
anteriorly, then
y=L(&) (3.1)
where L is the length of the longitudinal strips at 90" to the baseplate. For a 45"
arrangement this reduces to y = 1,. The ( x , z) coordinates within the transaxial
slice were centred on the midpoint of the localization frame, which again could
be established from the images of the thin aluminium strips. Some workers have
used three such frames, one either side and the other across the back of the head
for increased accuracy.
Greitz et a f (1980) showed many photographs of this system of fixation in
use with six different diagnostic units (two CT scanners, a PET tomograph, a
gamma camera and two x-ray systems)and three therapy units (a linac, the Leksell
multicobalt device and a stereotactic biopsy unit).
Stereotacticcoordinates are directly transferable from one 3D imaging modality
to another. The method is as applicable for the abnormal brain as for the normal
brain, unlike techniques (see later) which aim to correlate functional images with
anatomical images from a stereotactic atlas.
Zhang et a f (1990) used a commercially available stereotactic system made of
non-ferrous, non-electrically conducting material and aluminium. Plexiglas plates
were attached to the sides of the frame with a 'Z'-shaped channel pattem which

Introduction 139
Figure 3.4. The patient positioned in the CT scanner wearing the

immobilization device and the localizationframe, as developed by Bergstrom.
The base of the frame is attached to a special head-holder. (From Bergstrom
et a1 ( I 981).)
provided the spatial localization in the same way as the system of Bergstrom et a1
(198 1). The ‘Z’-shaped channel was filled with copper sulphate solution for h4R
imaging, with a 68Gapositron emitter for PET and with aluminium rods for x-ray
CT. These imaged to points in transaxial planes enabling the data to be registered.
The main feature of the development of Zhang et a1 (1990) was the construction of
a sophisticated computer display and analysis environment. This was VAX based,
used common image formats for data from each of the modalities, provided for
point and region-of-interest transfer from one set of images to another and control
via pull down menus. Multiple images were viewable simultaneously. Zhang et
a1 (1 990) claimed this to be a significant advantage over other systems developed
for assisting neurological stereotactic surgery.
One of the disadvantages of a system such as that of Bergstrom et a1 (1981)
is that the patient may be uncomfortable and may not be able to see or hear.
Some PET studies are carried out to investigate the response of the brain to
visual and auditory stimulae which would be difficult in such a frame. With
this in mind Mazziotta et a1 (1982) designed a head-holder providing a stable
enclosure, but not screwed to the skull. The anterior part of the holder was a pair of
hinged wings which provided an emergency escape in case of nausea or breathing
difficulties, and left the eyes and ears free. The rear of the head-holder was fixed
to the scanning couch. Reproducible positioning within the PET scanner was thus
achieved. Also correlated studies between x-ray CT and PET tomographic studies
were performed.
This system had no localization frame and a different method of registering
image data from the two tomographic modalities and with planar x-rays was used:
With the patient in the PET scanner the laser which localized the slice was tracked

Figure 3.5. The localization method of Meltzer et a l . The line ‘a’ defines the
transaxial plane in the MRI data-set. Line ‘b’defines the orientation of some
oblique plane of interest which intersects a particular (here the 4th) localizer
tube. When the patient was transferred to the PET scanner, the localizing laser
was pointed towards this marker, thus ensuring the PET transaxial planes were
parallel to the chosen MRI oblique planes. (From Meltzer et a1 (1990).)
on to the head-holder and marked with a waxed pencil. A thin aluminium wire was
then put in this place to mark the tomographic slice. This wire was clearly imaged
on plane x-rays and knowing the magnification of the x-rays and the interplane
tomographic spacing, all transaxial PET images were related to the anatomy as
depicted on the plane x-rays. By aligning the laser system of the CT scanner with
the wax mark, it was ensured that x-ray CT tomograms could be matched to the
corresponding PET data.
Miura et a1 (1988) developed a similar but not identical system for image
registration. The patient wore a mask with eye-holes which was not fixed to the
couch. The orbitomeatal line was marked with a lead string and used to position
the patient within the PET or x-ray CT scanner so the transaxial direction aligned
with this. Either side of the mask were two aluminium tubes at 45” to each other,
able to contain a thin steel wire or a tube of positron-emitting isotope. These
imaged to four points in each transaxial modality and CT-based anatomical images
were aligned with PET-based functional images by scaling, rotating and translating
one data-set relative to the other until the trapezia defined by the four points
overlapped, at which stage the data were registered and regions of interest were
transferable between them. The geometrical transformation was needed because
of the lack of tying the frame to the couch, unlike the technique of Mazziotta et a1
(1982).
Meltzer et a1 (1990) used a head-holder, not locked to the couch, to correlate
MRI, CT and PET data. The novel feature of this system was a ‘localizing bar’ fixed
centrally over the nose region containing 20 evenly spaced markers of petroleum
jelly. The markers showed as 20 bright dots on the mid-sagittal MRI image.
A particlular oblique slice of interest was then defined by measuring the angle
between the transaxial MRI plane and the oblique plane and finding which marker
this oblique plane intersected. When the patient was transferred to the PET scanner
the laser alignment system was directed to intercept this marker, thus assuring that
the PET transaxial slices were then parallel to the required oblique plane on the MRI

Introduction 141
Figure 3.6. The relationship between tomographic coordinates and stereo-

tactic coordinates, as used in atlases. The method of establishing this is de-
scribed in the text. 'ac' is the line between the anterior and posterior com-
missures (AC-PC line), 'gi' is the glabella-inion (GI) line (visible on a head
radiograph). The two lines are parallel and separated by 0.21 times the per-
pendicular distance from the GI line to the vertex of the skull. '0'is the centre
of the tomographic data, 'e' is the centre of the stereotactic coordinateflame.
Seven tomographic slices are shown (fromthe PETi-6 scanner). X" is the an-
gle between the GI line and the transaxial direction. The brain size can be
measured from the radiograph (length of AC-PC line and distance Z,) and
corrected for magnification to provide the scale factor between the real brain
and the atlas. (From Fox et a1 (1985).)
images (figure 3.5).

Fox et a1 (1985) used a head-holder (comprising a lucite hemicylinder, shape-
assuming foam and a plastic face mask) during PET scanning with the PE'IT6
emission tomograph. The aim of their work was to correlate PET images with
images from a stereotactic atlas rather than patient-specific CT or MRI images.
A lucite plate, this time containing seven wires at the interslice spacing of the
tomograph, was mounted on the head-holder. The wall-mounted laser, defining the
first tomographic slice, was aligned with the first wire and a plane x-radiograph
was taken showing all seven wires defining the tomographic slices. The origin
of the tomographic data was taken to be the middle pixel of the fourth (i.e.
middle) slice. In order to interpret the PET data, Fox et a1 (1985) established
the relationship between this coordinate system and the stereotactic coordinate
system used by neurological atlases (figure 3.6). The origin of the stereotactic
coordinate system was the mid-point of the (AC-PC) line between anterior and
posterior commissures of the brain in the mid-sagittal plane of the brain. Since
this line cannot be seen on a plane x-ray image, it was drawn parallel to the
glabella-inion (GI) line connecting the most anterior point of the glabella (g) to
the most posterior point of the base of the inion (i) at a distance of 0.21 times the
perpendicular distance (Z),, from the GI line to the skull vertex (p). The three
assumptions that the GI line can be accurately and reproducibly found from a
plane x-ray image, that the GI line is parallel to the AC-PC line and at this precise
distance, were extensively validated by Fox et a1 (1985). The angle between the

AC-PC line and the transaxial direction can then be established. It then becomes a
relatively simple matter to relate the two sets of coordinates via this angle and the
three translations between the origins. Specific measures were taken to ensure no
other rotations occurred. All that remained was to scale the recorded tomographic
data to the scale of the atlas.
Fox et a1 (1 985) demonstrated the usefulness of their technique by sucessfully
locating in the stereotactic atlas the anatomic features responsible for observed and
measured functional response, in particular the cavemous sinuses (associated with
high local cerebral blood volume) and the visually responsive cortex (associated
with increased regional cerebral blood flow).
The main limitation of atlas-based correlations with functional images is that
the method is really only valid for normal anatomy. Several large atlases of
the anatomy of the brain are useful for this purpose, including Talairach and
Szikla (1967), Matsui and Hirano (1978), Hanaway et a1 (1980) and Talairach
and Toumoux (1988).
Wilson and Mountz (1989) have developed a frame for registering functional
and anatomical images which overcomes some limitations of the others. The
frame comprises two equilateral triangles of nylon carrying thin tubing in which
x-ray attenuating material, positron emitting isotope or nickel chloride fluid can be
placed for CT, PET and MR imaging respectively. There were two such triangular
frames and they were positioned either side of the head and located firmly using
stethoscope earplugs in the auditory meatus and fixation to the lateral canthus.
Each frame also had a nylon bar with tubing running from base to vertex (figure
3.7). The contrast materials imaged to six bright dots in each tomographic plane,
three either side of the transaxial head image. The angle 0 of the image plane to
the base of the triangles is a simple function of the ratio A / B , where A and B are
the distances between the point images in that plane, i.e.
The 'level' L of the image plane (where the plane meets the vertical from the left-
hand comer of the triangle) depends again on A and B and on 'f' (the half-side of
the equilateral triangle) via
( t - B COS e ) 2~ tan 60"

L= (3.3)
(B +A)
So from simple measurements and these formulae, the tomographic plane can be
specified. By aligning the images from different modalities such that the six point
images coincided, the images were registered. The algorithm for transferring one
3D image set into the frame of the other was driven by the images of the planes
in which the dots were furthest apart for accuracy (i.e. the most inferior planes
towards the base of the triangles).

Introduction 143
IMAGE
PLANE ANT,
TI
POST. ANT. POST.
Figure 3.7. One of the triangular frames in the method of Wilson and
Mountz, showing the equilateral triangle with the line from the vertex to the
base. Any particular image plane intersects this line and the two sides of the
triangle in three points (as shown to the right). By measuring the distances
A and B on the right-hand diagram, the angle 9 and the ‘level’ L of this
tomographic plane may be determined by equations (3.2) and (3.3). (From
Wilson and Mountz (1989).)
A relocatable stereotactic frame comprising a baseplate to which is secured a

dental impression of the patient’s upper jaw has been designed by Gill (1987). The
baseplate and dental plate are secured to the head with adjustable straps. Together
with an individually moulded occipital impression, this provides a non-invasive
system of support. The dental impression is taken in polyethylmethacrylate
polymer which sets hard in afew minutes with negligible shrinkage. The straps are
of woven nylon with buckles of plastic to avoid artefacts when imaging. The head
support is hard-setting rubber and the baseplate is made of epoxy. The accuracy of
positioning is 1 mm (Graham et a1 1991b), significantly better than moulded head
casts. The frame can be used with a Brown-Roberts-Wells (BRW) stereotactic
frame (Heilbrun et al 1983) attached for positional localization (Thomson et a1
1990, Thomas et a1 1990, Gill et a1 1991) in both diagnostic imaging units and at
the time of therapy.
A stereotactic frame such as that described by Brown (1979) essentially pro-
vides a means of relating three-dimensional image coordinates to correspond-
ing three-dimensional locations in frame coordinates. This is achieved by imag-
ing the points at which the components of one or more frame-constituent ‘Z’ or
“’-shaped tubes (whose locations in frame-space are precisely known) cross the
transaxial CT slices, giving characteristicallyelliptical-shapeddots. The measured
coordinates of these points in the two spaces determines the matrix transformation
which then allows any points in one space to be known in the other space. Under
these conditions, the stereotactic frame need not be attached to the imaging device
or its couch, provided it is reproducibly attached to the patient’s skull.
One of the main uses of this arrangement was to determine, on the CT data, the
target point for a rigid intracranial probe and its entry point into the brain and thus
specify where to insert the probe in frame coordinates. The computer can then
indicate the point of intersection of this probe with each transaxial CT slice and
if the probe trajectory were unsatisfactory a new one could be determined. In the
context of radiosurgery (see section 3.2) it would equally be possible to investigate
the ‘trajectory’ of a radiation beam in the same way.

3.2. RADIOSURGERY AND STEREOTACTIC RADIOTHERAPY
Historically the term ‘radiosurgery’ was introduced by Leksell(l95 1) to describe

the method by which many narrow radiation beams were used to irradiate
intracranial structures from many angles as an altemative to surgery. For many
years 6oCo sources were used and commercial units were produced. Charged-
particle irradiations (especially protons) have been used, but facilities are very
expensive. Adapted CT scanners and radioactive implants have also been
used. More recently, isocentric linear accelerators have taken over this role and
corresponding interest in the technique has greatly increased.
The terms stereotactic radiosurgery and stereotactic radiotherapy should be
distinguished; the former is applicable to a single-fraction treatment equivalent
to a surgical operation, whereas the latter involves therapy by multiple fractions.
Radiosurgery is considered for intracranial lesions considered inoperable or
carrying excessive risk from conventional surgery. Unlike conventional
radiotherapy, the lesion is not ‘treated’ so much as ‘destroyed’. Indeed, as
well as applications in cancer medicine, one of the commonest applications
of radiosurgery is the treatment of benign arteriovenous malformations (AVM).
Stereotactic radiosurgery obliterates tissue by inducing gliosis or fibrosis within it.
The therapeutic principle is quite unlike radiation therapy, which aims to preserve
healthy cells within the target volume. Radiosurgery is appropriate for single-shot
small-volume irradiation where precision is meticulously sought and high dose
gradients are required.
Barrow et a1 (1990) point out however that the vast majority of neurological
lesions are best treated by conventional surgery and that the indications and
contra-indications for radiosurgery must be considered very carefully before this
procedure is used. It is not expected to replace surgery and its role must be kept
in perspective. Barrow et a1 (1 990) sum up many of the clinical arguments. The
physicist might remark that it is unlikely that radiosurgery, important though it is,
would ever provide the unique justification for purchasing a cyclotron or a linac
and this may explain why the special purpose devices (see section 3.2.1) are also
so rare.
Whatever radiation-based treatment is selected, precise knowledge of the
location of the target is essential. Techniques for registering 3D MRI, CT and
PET data with digital subtraction angiograms have been developed by Henri er
a1 (1 989). Bova and Friedman (1991) have explained why biplane angiography
alone might be less than satisfactory. Herbert and Frijder (1990) have developed
digital tumour fluoroscopy (DTF),analogous to digital subtraction angiography,
whereby an x-ray generator and image intensifier are used to create images in the
operating theatre itself. Localization requires complete tomographic datasets in
general. All radiosurgery involves the use of some kind of immobilization and
localization frame of the type discussed above. When such a stereotactic frame is
screwed to the skull it is used in the ‘target-centred’ mode (as opposed to ‘frame-
centred mode’-see Galloway and Maciunas 1 9 9 0 t t h a t is, the target is placed at

Radiosurgery and stereotactic radiotherapy 145
the centre of the pointing apparatus and all trajectories reach the target regardless
of orientation.
3.2.1. Radiosurgery with 6oCosources; the gamma knife

Apart from the early use of protons for radiosurgery (Larsson et a1 1958) and
Leksell's original apparatus, which had an x-ray set attached to a stereotacticframe
(Leksell 195l), historically the technique has relied on 6oCoradiation (Leksell
1971, Goodman 1990, Barrow 1990). The first so-called 'gamma knife' (figure
3.8) was constructed by AB Motala Verkstad, Motala and installed at Sweden's
Sophiahemmet Hospital in 1968. It was later moved to UCLA School of Medicine
(Leksell 1971, To et a1 1990). The Karolinska Hospital in Stockholm developed
the second gamma knife and its arrival in 1975 coincided with the widespread
use of computed tomography, which gave impetus through improved accuracy to
the technique of stereotactic radiosurgery (Leksell 1987). To date there have only
been nine installations in the world (Maitz et a1 1990, Goodman 1990) (see table
3.1).
The first gamma knife had 179 sources; later models have 201. The Leksell
201 source 6oCogamma knife is shown in figure 3.9 as installed at the facility at
the University of Pittsburgh, Pennsylvania (Wu et a1 1990). The 201 sources are
contained in the upper hemispherical shield. All sources are focused to a single
point at a source-to-focus distance of 403 mm. The central beam is at 55" to the
horizontal plane. The sources lie in an arc of k48"about this central ray in the
long axis of the treatment unit and at f80"along the transverse axis to the patient
couch, The sources are contained in two hemispherical shields made of iron. The
slideable entrance door is of steel. The whole unit weighs some 18 000 kg. The
total source actiity was 219 TBq in July 1987 delivering 398 cGy min-I. Unlike
previous units, which had been installed to their hospital sites with the sources
in position, the unit at Pittsburgh was loaded on site for reasons and by methods
discussed by Maitz et a1 (1990). Electrically generated hydraulic pressure is used
to move the treatment couch and to close the shielding door. In the event of failure
several back-up systems are provided.
The final collimation is achieved by means of one of four interchangeable
collimator helmets with 201 channels aligning with the sources and removable
apertures that produce 4, 8, 14 or 18 mm diameter fields at the focus: 4 mm is
probably the smallest beam any machine can produce (Wu et a1 1991). On the
other hand, the 18 mm size of the largest collimator hole limits the usefulness of
the device for irradiating large tumours. Individual plugs can stop-off selected
holes if required to shield the eyes or to change the dose distribution. Flickinger
et a1 (1990b) show suggestions for which holes to block and the effects which
this has on the three-dimensional dose distribution. Since the dose gradients are
very high with the gamma knife, precise conformation of the high-dose region to
a possibly irregularly shaped target volume is important.
A stereotactic frame is attached to the patient under local anaesthetic and in

Figure 3.8. ( a )Thefirst Leksell ‘gamma knife’. The sources are contained
in the domed structure to the right of the picture. The collimating helmet
can he seen on the couch next to the two staff. (6) Positioning the patient in
the helmet for the Leksell gamma knife. (From Leksell(1971).) (Courtesyof
Charles C Thomas, Publisher,Springfield, Illinois, USA.)
tum the frame is positioned into the collimating helmet so that the tumour to be
treated is at the focus. To begin treatment, the couch with the patient so positioned
is hydraulically driven into the gamma knife unit and the helmet mates with the
primary collimation. After a prescribed time the couch reverses out. Typically the
unit can deliver 300-400 cGy min-’, so several minutes treatment time is required
for a ‘singleshot’ (Coffey eta1 1990,1991,Kondziolkaetall990, Kondziolkaand
Lunsford 1991, Lunsford et a1 1990a,b). If the lesion is large, several ‘shots’ (i.e.
treatments with the focus at different brain locations and possibly with different
helmets) may be necessary.

Table 3.1. Gamma knife facilities worldwide (data from

Maitz (1990) and Goodman (1990)).
Date Installation
1968 Sophiahemmet Hospital, Stockholm, Sweden

1975 Karolinska Hospital, Stockholm, Sweden
(1981 1st Stockholm device donated to UCLA)
1983 Buenos Aires, Argentina
1985 Sheffield, England
1987 University of Pittsburgh, School of Medicine, PA
1989 Charlottesville, Virginia
1989 Chicago, Illinois
1989 Piedmont Hospital, Atlanta, Georgia
1991 Lawrence Livermore National Laboratory
Collimator,201 p c s
Beam sources, 201 pcs. A-{ Upper hemispherical shield
i
--I_
Cenlral body. -
die power
Base shield-
Figure 3.9. Schematic of the Leksell 'gamma knife' unit usedfor stereotactic
radiosurgery (see text for details). (From Maitz et a1 ( I 990).) (Reprinted with
permission from Pergamon Press Ltd. Odord, U K . )
The gamma knife at Sheffield, UK, constructed by Nucletec SA (Switzerland),

is similar (but not identical), comprising 201 sources of total activity 209 TBq
distributed over a spherical sector of 60" by 160" with a source-to-focus distance
of 395 mm. There are three helmets whose holes are 4, 8 and 14 mm diameter
(Walton et a1 1987). Dose distributions for each helmet were computed by
integrating over the sources using the measured dose profile transverse to each
beam. They were also measured and found to be in good agreement. With the
14 mm helmet assigned an output factor of unity, giving 295 cGy min-' in Sept
1985, the output factors of the 8 mm and 4 mm helmets were 0.964 and 0.925,
respectively. The unit is used as described above for the Pittsburgh unit.
The main disadvantages of the gamma knife are that it cannot be 'tumed off'
and once the shield is open, the patient receives whole-body dose before the helmet
locks into the primary collimator. The cobalt needs replenishing periodically and

PlTT GAMMA KNIFE
Single -am profilas wtIh background at Bcm
deplh In spherlcdl phanlom
1
8- mm COLLIMATOR
4
b- Bmm COLLIMATOR
c- 14" COLLIMATOR
a - 18" COLLIMATOR
e- BACKGROUND
i1
i
I
:;,
@O
,
4
,
6
,
12
\
, , ,
,
16
RADIAL DISTANCE FROM BEAM AXIS (mm)
20
,
24
1
Figure 3.10. Raw dose profiles at 8 cm depth of a single beam from the
Pitt gamma knife, measured with all but one of the 201-collimator-holes
collimator plugged. Due to the transmission through the 200 collimator
shieldings, the profiles show a considerable background, which appears in the
form of broad shoulders on either side of the peak. Thefour curves correspond
to the different helmets (collimators). (From Wu et a1 ( I 990).) (Reprinted with
permission from Pergamon Press Ltd, Oxford, UK.)
the unit is very expensive ($3.5 million according to Barrow 1990).

Flickinger and Steiner (1990) and Flickinger et a1 (1991) discuss the dose-
response predictions for stereotactic radiosurgery with different collimation
helmets.
Dose planning for treatments with the Leksell gamma knife requires knowledge
of the dose profile from a single beam. Wu et a1 (1990) measured the dose profile
by blocking off 200 of the 201 sources. The raw profiles obtained (figure 3.10)
display a substantial shoulder representing almost 60% of the maximum intensity
of a single beam. This is due to the background radiation from the plugged sources.
The background was measured by plugging all 201 sources and subtracted from
the profiles in figure 3.10 to give the 'true' single-beam dose profiles, shown in
figure 3.11, for the various collimators. The shape of these gives the off-axis ratio
(OAR) needed for planning.
Figure 3.12 shows how the treatment planning program computes the dose to
a point P in the patient. The manufacturer specified the dose rate at the reference
point R, 1 cm deep in a spherical phantom of radius 8 cm, i.e. 7 cm from the focus
of all the sources. Hence the distance SR is 333 mm. The on-axis dose rate at
the point Q, (403 - z) mm from the source, can be established from the inverse-
square law and the linear exponential attenuation. Finally, the dose-rate at an off-
axis point P follows by applying the OAR from the dose profile. This procedure
is repeated for all sources and the results summed to give a three-dimensional
dose distribution on a 313 dose matrix spanning the target volume. The necessary
measurements of source-to-skull distance (which affect the exponential term in
the calculation) are made by attaching a special plastic helmet to the baseplate of

PlTT GAMMA KNIFE
Single beam prollier af 8 Em In polystyrene
spherical phantom
COLLIMATOR HELMET
80 a- rmm
b- 8"
"r it I
1
ib
-RADIAL DISTANCE FROM BEAM AXIS (mm)
Figure 3.11. 'True' dose profiles for single beams in the Pitt gamma knife
for four different collimator sizes. These sizes are defined by the full-width
at half-maximum of the SO% isodose line. These curves are derived from the
data in figure 3.10. (From Wu et a1 (1990).) (Reprinted with permission from
Pergamon Press Ltd, Odord, UK.)
5
Figure 3.12. A schematic diagram of a simple dose calculation algorithm

developed by the manufacturers of the gamma knife. A dose reference point R
was chosen 1 cm from the surface of an 8 cm spherical phantom. The dose at
any general point Q in the calculation matrix is then obtained using the dose
profilesfrom figure 3.11 and the inverse-square law and the linear-attenuation
exponential formula. (From Wu et al ( I 9 9 0 ) ~(Reprinted with permission
from Pergamon Press Ltd, Odord, UK.)
the stereotactic frame and using a 'dip-stick' at selected locations on the surface of
the helmet to make measurements from which the computer calculates the three-
dimensional surface of the skull for treatment planning.
Some gamma knife facilities do not have the 18 mm diameter collimated helmet,
restricting the range of size of targets. Even the 18 mm collimator is not large
enough for some treatments. The problem can be overcome by treating with two
or more isocentres displaced from each other. This generates non-spherical dose

1 I
0.0 0.5 1. o 1.5 2.0
SEPARATION (cm)
Figure 3.13. Showing the change in the diameter of the volume defined
by the 50% isodose in three orthogonal directions as a function of the
separation between isocentres for a radiosurgery gamma-knife unit with an
18 mm diameter collimator-helmet. The isocentres are separated along the
x-direction. (From Flickinger et a1 (1990a).) (Reprinted with permission
from Pergamon Press Ltd, Oxford, UK.)
distributions. Flickinger er a1 (1990a) plotted the shape of the distributions for

two-isocentre treatments as the separation between the isocentres was increased.
For example, if the isocentres were separated along the x-direction the diameter
of the volume enclosed by the 50% isodose line expanded in the x-direction and
contracted in the orthogonal directions with increasing separation (figure 3.13).
Flickinger et a1 (1990a) computed a catalogue of 3D dose distributions with
multiple isocentres of varying separation in order to guide the treatment planning
process. Each plan was customized from this start.
Some targets could be planned with one- or two-isocentre treatments adequately
encompassing the target within the 50% isodose surface. The problem then arose
of which was the better treatment. Flickinger et a1 (1990a) generated dose-volume
histograms in surrounding normal brain, showing the improvement obtained with
two isocentres and used a biological model to show the NTCP was reduced,
justifying the increased effort (figure 3.14).
The method of using multiple isocentres to treat larger brain lesions with the
gamma knife has many similarities with the corresponding technique for multiple-
arc radiotherapy with multiple isocentres (see section 3.2.4).
3.2.2. Stereotactic multiple-arc radiotherapy with a linear accelerator

The isocentrically mounted linear accelerator is a natural choice for the x-
ray source for radiosurgery. Whilst an x-ray source was Leksell’s first choice
for stereotactic radiosurgery (Leksell 1951), it was abandoned because of the
difficulties in aligning the beam. The move towards using linear accelerators is a
consequence of the much improved bearings (Barrow er a1 1990, Podgorsak er a1
1989) of modem accelerators. All the experience from the gamma knife is relevant

------ol
10 I 'L -
- Brain 14"
Brain 18 m m
- 1000
- 100
W
'1
I
3 I
2 .1 j
9 i
.O1 i L.
001 J
Yo I
01
I
,0001
0 20 4 0 6 0 8 0 100
*iISODOSE
o
Figure 3.14. Dose-volume histograms for normal brain tissue comparing

treatments of a 21 mm by 14 mm by 14" tumour with either a 18 m m
collimator and a single isocentre or a 14 mm collimator with two isocentres
separated by 9 mm. The latter is to be preferred, verified by a biological model
giving the NTCP. (From Flickinger et a1 (1990a)J (Reprinted with permission
from Pergamon Press Ltd, Oxford, OK.)
since there is no reason to believe these different systems will produce different
outcomes in similar groups of patients, Further advantages include the lower cost
of a linac and its availability for 'conventional' radiotherapy when not used for
radiosurgery. Betti and Derechinsky (1982) and Heifetz et a1 (1984) were among
the first to study the potential, including making the suggestion that multiple arcs
could be made by rotating the couch to different positions. Many authors have
since discussed linac-based arc therapy including: Betti and Derechinsky (1982),
Betti et a1 (1989), Hartmann et a1 (1985), Colombo et a1 (1985a,b, 1987, 1990),
Patil(l985, 1989), van Buren etal (1986), Chierego etal (1988), Harizetal(l988,
1990), Lutz et a1 (1988), Saunders et a1 (1988), Friedman and Bova (1989), Pastyr
et a1 (1989), Casentini et al(1990), Croft (1990), Delannes etal (1990), Engenhart
et a1 (1990), Loeffler et a1 (1 990a,b), Schwade et a1 (1990), Thomson et a1 (1990),
Grahametal(1991a),Gill eta1(1991), Schell eta1(1991),andSeragoetal(1991).
By arranging for the centre of the tumour to be at the isocentre, treatment
becomes possible using the beam from a number of non-coplanar arcs by
combining gantry rotation with a finite number of couch positions with twist.
Fractionated radiotherapy also becomes possible, potentially increasing the
therapeutic ratio (Graham et a1 1991a). The success of fractionated radiotherapy
depends on the reproducibility of patient position, both between diagnostic
imaging sessions and between the treatment fractions. This generally demands
use of a reliable head-holder and stereotactic frame, although Heifetz et a1 (1989)
suggest that using external scalp landmarks may be acceptable for large-volume
irradiations. Thomson (1990) and Thomson et a1 (1990) used the Gill frame for
localization and five arcs of 140" of rotation. A Clinac 6/100 6 MV treatment unit
was equipped with a special applicator to minimize penumbra. Planning software
was written in house (Thomson 1990). They proposed the replacement of the term

‘radiosurgery’ by ‘stereotactic multiple-arc radiotherapy (SMART)’.

Gill et a1 (199 1) report that, in their use of a prototype head-frame constructed
with a dental impression, occipital support and straps and with a BRW frame
for external beam radiotherapy, the positioning uncertainty was less than 1 mm.
This allowed fractionated radiotherapy to be confidently practised. Dose planning
was performed with an IGE TARGET system. Because the dental impression
acted only on the upper jaw, the patient was comfortable and free to talk. The
securing straps were of woven nylon, which produced no artefacts in diagnostic
imaging. The straps were secured with a plastic buckle, which gave a quick-
release mechanism in case of emergency.
Hartmann et a1 (1985) and Pastyr et a1 (1989) developed the ‘Heidelberg
system’, based on the use of a Riechert-Mundinger head-frame and 11 rotational
non-coplanar arcs. Friedman and Bova (1989) developed a linac-based
radiosurgery facility in which an extra mechanical system was provided which
functioned independently of the proprietry rotation bearings for couch and gantry.
This enabled an accuracy of 0.2 f 0.1 mm to be achieved, comparable with the
accuracy of a gamma knife. An in-house-developed planning system also allowed
almost real-time visualization of dose contours relative to anatomy.
Barrow et a1 (1990)and McGinley et a1 (1990) described the ‘Emory x-ray
knife’ at the Emory Clinic, Atlanta, Georgia, USA (figure 3.15). This utilized
a BRW immobilization scheme, with the patient seated upright on a rotatable
tumtable. The centre of the lesion being treated was arranged to coincide with the
centre of rotation of the turntable and also with the axis of rotation of a 6 MV linear
accelerator fitted with a circular collimator to define a small beam of diameter 5
to 30 mm. The gantry could take up angles from 35” to 80” to the vertical. The
gantry did not move during irradiation, thus reducing the risks of collisions in the
event of electrical failure. During irradiation, the patient turntable rotated slowly
at 3-4 RPM. The dose rate of the accelerator was sufficiently low to allow at least
four complete rotations of the tumtable at each gantry angle. By using several
gantry angles with rotation at each angle, near-spherical high-dose volumes were
created. Experiment showed (McGinley et a1 1990) that the 90% isodose was
approximately 4 mm smaller than the collimator diameter and the 80% isodose
was 2 m m smaller than the collimator diameter. The shape and size of the high-
dose region was independent of position within the brain. Non-spherical dose
distributions could be obtained by differentially weighting the number of monitor
units for each gantry angle.
Stereotactic radiosurgery with a linear accelerator requires a special-purpose
collimator to reduce penumbra and so sharpen the dose gradient at the edge of the
target volume. Circular collimators were described by Hartmann et a1 (1985) of
size 5 mm to 20 mm diameter for use in Heidelberg. Croft (1990) shows circular
collimators of 12.5 mm to 30 mm diameter. Figure 3.16 shows one such collimator
designed by Winston and Lutz (1988) for the Boston system, which generally uses
four arcs. The circular assembly extends to within 23 cm of the isocentre and will
accept a range of eight different cerrobend inserts with different conically drilled

Figure 3.15. The Emory x-ray knife. The patient is seated on a rotating
turntable. The linear accelerator gantry rotates about an axis coinciding with
that of the turntable. The intersection point is the centre of the brain lesion.
The gantry is stationary during irradiation. Notice the special collimator.
(From Barrow et ai ( I 9 9 0 ) ~
openings, whose diameters, projected to the isocentre, range from 12.5 to 30 mm

in steps of 2.5 mm (Lutz et al 1988). Figure 3.17 shows the static beam profiles
for three diameters of collimator measured experimentally. When several arcs
of radiation are combined, the three-dimensional dose distributions show that the
volume of non-target tissue receiving a high amount of radiation increases rapidly
as the diameter of the collimator increases, but nevertheless is confined close to
the target volume (for detailed calculations see Winston and Lutz 1988).
The potential of linear-accelerator-based radiosurgery depends on accurate
verification. The delivery of single shots of very high dose to small volumes of
brain tissue will be intolerant of error. For this reason radiographic verification
must be performed to ascertain that the collimator is accurately positioned relative
to the target, at all orientations of the gantry. Each development should be

Figure 3.16. ( a )Linear-accelerator-based stereotactic radiosurgery, show-

ing the coincidence of the axes of rotation of the gantry (G),the turntable (T)
and the collimator (C) at the isocentre of the machine. The radiation col-
limator will extend to close to the isocentre. (b) The small-field collimator
supplements the internal rectangular collimator. The aperture of the collima-
tor can be adjusted by inserting different cerrobend castings. Each collimator
is labelled by the diameter of the field at the isocentre. (From Winston and
Lutz (1988).)
Figure 3.17. Static beam profiles for three circular collimators used for
stereotactic radiosurgery. Notice the very rapid decline at the edge of the
field and the nearly identical shape of these edges. (From Winston and Lutz
( I 988).)

critically assessed with this in mind. Winston and Lutz (1988), Lutz et a1
(1988) and Saunders et a1 (1988), reporting the Boston system, have provided
a particularly rigorous technique incorporating verification and error assessment.
Tsai et a1 (1991) have discussed extensively the Boston quality assurance
programme, which underpins the safety of the stereotactic irradiation technique.
Drzymala (1991) also addresses issues of quality assurance. Amdt (1989),
however, raises doubts concerning the accuracy of linac-based radiosurgery
compared with the accuracy of the gamma knife, whose errors are smaller than
0.3 mm.
Hartmann et a1 (1985) used a dose computation algorithm based on integration
of superpositions of beams whose radial and depth-dose profiles had been
measured experimentally. Dose distributions were computed in orthogonal planes
through the target volume. Consideration of integral dose to the normal brain
led Hartmann er a1 (1985) to conclude that stereotactic radiosurgery should be
confined to tumours of 3 cm diameter or less. This is now a commonly accepted
limitation for single-shot irradiations. Heifetz et a1 (1989) suggested that larger
volumes could be treated by fractionated radiosurgery. Multiple-arc techniques
have also been applied to stereotactic radiotherapy of head and neck lesions
(Hartmann et a1 1987).
For those implementations of the method in which the head is held in a frame
fixed to the floor, it is necessary to deactivate the motors which drive the couch
to protect the patient during therapy (Croft 1990, Lutz et a1 1988). The total time
required for treatment is a function of the output of the linear accelerator in monitor
units (MU) per degree, and can typically be some 0.5-0.75 hours (Croft 1990).
However, the time taken to attach the frame, to take 3D images, perform planning
and set up the treatment can easily be a whole day. Stereotactic radiosurgery is
very time consuming.
3.2.2.1. Dynamic stereotactic radiosurgery. If a single transaxial arc is used,

the dose gradient in the plane of the arc will not be great because, for a full 2n
rotation (see e.g. chapter 2), each beam has a parallel-opposed counterpart. The
dose gradient in the orthogonal direction, normal to the plane of rotation, will
be as good as for the static beam. Multiple arcs help to overcome this problem.
An alternative solution was provided by Pike et a1 (1987) and Podgorsak et a1
(1987, 1988, 1989), called 'dynamic stereotactic radiosurgery'. The patient was
positioned isocentrically so that the axis of the gantry and the axis of couch rotation
passed through the centre of the spherical target volume. The gantry was rotated
from 30" to 330" whilst the couch simultaneously rotated from -75" to +75" with
the beam switched on. The entrance trajectory thus describes a complex curve on
the patient's head in which the beam always enters from the upper hemisphere and
parallel-opposed fields do not occur. The situation is summarized in figure 3.18.
Podgorsak et a1 (1988) showed that this leads to dose gradients which improve
upon those from single-arc rotations for some planes of measurement, but are
worse for others. When a comparison is made with the dose profile for a single

GAMMA UNIT LINEAR ACCELERATOR LINEAR ACCELERATOR LINEAR ACCELERATOR

(201 COBALT BEAMS) (SINGLE PLANE ROTATION) NONCDPUNAR (DYNAMIC ROTATION)
CONVERGING A m
Figure 3.18. A comparison of points of beam entry into the patient's skull
for various competing radiosurgical techniques. ( a )The Leksell gamma knife,
( b ) the single transaxial plane rotation with a linear accelerator (simple
rotation therapy with a very small area beam), (c) a multiple non-coplanar
converging-arc technique with a linear accelerator, ( d )dynamic stereotactic
radiosurgery with a linear accelerator. (From Podgorsak et a1 (I989).)
(Reprinted with permission from Pergamon Press Ltd, Odord, UK.)
arc in the plane of its rotation, dynamic radiosurgery is always superior, whatever
the radius of the beam. Dynamic radiosurgery cannot, however, improve the ratio
of integral (brain) dose to the target dose.
One of the major disadvantages of using the complicated entrance trajectory is
that treatment planning becomes very difficult and Podgorsak et a1 (1988) derive
the target dose even for this trajectory from a single-arc calculation. This does
not get the correct dose distribution outside the target volume so this is estimated
from profiles of dose gradients.
Podgorsak et a1 (1988) use an in-house-constructed stereotactic frame with 3D
image-based dose calculations. Corrections are made for the photon attenuation
in the frame. Circular collimators shape the field. The clinical experience at the
McGill University facility has been summarized by Souhami et a1 (1990a,b).
A detailed comparison of the dose fall-off outside the target volume for the
gamma knife, the single-plane rotation, the Heidelberg multiple-converging-
arc technique, the Boston multiple-converging-arc-technique and the McGill
University dynamic stereosurgery method has been provided by Podgorsak
(1989). Although measurements of dose gradient are not true volumetric
assessments (Graham et a1 1991a), this information (see figure 3.19) leads to the
conclusion that the dynamic stereosurgery technique can rival the gamma knife,
which is seen as a gold standard. Since none of the methods for linac-based
radiosurgery provide a uniform distribution of entry points over the surface of the
skull, all dose distributions are slightly spherically asymmetric. Since the dose
gradient thus depends on the plane of measurement through the target volume,
figure 3.19 displays the best and worst gradients obtained with each method. All
other gradients lie within the extremes.

80
70
(I)
60
Dynamic rotation
30
20
, .p&=<$..l
- -
10
...--I I , ,
-
I
0 10 20 30 40 50
Distance from lsocenter (mm)
Figure 3.19. A comparison of dose gradient for diferent techniques for

stereotactic radiosurgery with a I cm diameter spherical target covered by the
90% isodose surface. Since all techniques provide an anisotropic distribution
of entry portals on the surface of the patient's skull, the dose distributions are
asymmetric and dose profiles depend on the chosen measurement plane. In
this figure the steepest and the shallowest profiles are shown for each method:
(a)for the gamma knife obtained from Walton et a1 (1987), and (b),(c), (d),
and ( e )for a 10 MV linear accelerator; (b) single plane rotation, ( c ) I 1
multiple converging arcs (Heidelberg system), (d)four multiple converging
arcs (Boston system),and ( e )the dynamic stereotactic radiosurgery technique
(McCill). (From Podgorsak et al (19 8 9 ) ~(Reprinted with permission from
Pergamon Press Ltd, Oxford, UK.)
3.2.2.2. The dose calculation algorithm in linac-based stereotactic radio-

surgery. Kubsad et a1 (1990) used the EGS4 Monte Carlo code to compute the
beam profile for a linear accelerator fitted with a circular collimator. Profiles were
computed by a full-scale simulation of particle transport through all the machine
components (primary collimation, flattening filter, secondary collimation, mov-
ing jaws, and stereotactic collimator). The same profiles were also independently
computed by the convolution method (equation (2.50)) and shown to be in good
agreement with the full calculation (and with experimental measurements). The
dose, computed this way, is
DE (r) = J J J TE (s) h ( E , r - S) d3s (3.4)
where TE is the terma (the product of the mass attenuation coefficient p / p ( E , r )

and the energy fluence E ( r ) )and h is the convolution kernel.
The dose distribution in multiple-arc linac-based stereotactic radiosurgery is
constructed as the superposition of the doses from each arc. In turn, each arc is
represented as a sum of fixed 'stationary' beams. Because these beams are small,

Figure 3.20. Conceptual diagram showing the dose calculation for a single
beam as part of a stereotactic multiple-arc therapy. The dose is calculated at
point P.The x-ray source is at X and the isocentre is at 0.All other quantities
are defined in the text. (From Luxton et a1 (1991).)
there is no need to correct for the shape of the patient contour when computing
off-axis points; this contour can be considered locally flat. Luxton et a1 (1991)
provide the formalism needed for each of these elemental beams for each arc:
0 Let the beam monitor be calibrated so the dose per monitor unit (MU) is 1 cGy
+
at a depth dmaxin water for a lox 10 cm field at a distance of 1 m dmm from
the source.
0 With the circular collimator in place, let O F ( S ) be the equivalent quantity for
a field of diameter S projected to the isocentre. This is called the output factor.
0 Let T M R ( S , d ) be the tissue-maximum ratio at depth d , field diameter S.
Then the single-beam dose per monitor unit to a point P a depth d from the
surface and distance r from the central axis of the beam is given by
D ( d , r , S , diso) = T M R ( S , d ) 0 F ( S ) R (r') ( S A D + /b2 (3.5)
where diso is the depth of the isocentre 0, SAD is the source-to-isocentre

+
distance and b = (SAD - dis0 d ) is the distance from the x-ray source to the
perpendicular onto the beam axis from the point at which the dose is calculated.
r' is related to r via
r' = r S A D / b - K (d - dmm) (3.6)
(see figure 3.20) and K is a small constant which accounts for scatter, the first term
being the correction for beam divergence. R (r') is the off-axis ratio at distance r'
from the central ray at a depth of dmaxin the plane of the isocentre and this may
be found by measurement of the profile.

Luxton et a1 (1991) measured beam profiles at the isocentre for a number of

depths in scattering material. As the depth increased the profiles broadened by a
small but measurable amount, allowing calculation of the parameter K by fitting
the data. Experimental measurements of the TMR showed that this could be
represented in the region d 2 d,,, by an analytic function
T M R ( d ) = exp ( - p (d - d")) (3.7)
where the coefficient p depended on beam energy and collimator diameter. They
also gave a formula for calculating the TMR for d <
dmm, but this is not an
issue in stereotactic radiosurgery. Luxton et a1 (1991) verified the accuracy of
the calculation algorithm by performing TLD and film-dose measurements for
irradiations of test phantoms. Gehring et a1 (1991) discuss a planning system in
which the weights of the arcs may be adjusted.
3.2.2.3. Dynamic collimation for stereotactic radiosurgery with multiple arcs.

The limitations of single-isocentre multiple-arc radiotherapy/surgery have
already been discussed. In order to generate non-spherical dose distributions,
several workers have developed multiple-arc therapy with multiple isocentres.
However, this can lead to difficulties in obtaining a uniform distribution of dose
within the target.
Leavitt et a1 (1 991) have presented a very elegant method of obtaining a uniform
dose to an irregularly shaped and possibly quite large target in the brain with
multiple-arc therapy by dynamic field shaping. Figure 3.21 shows how they added
four independent block collimators to the usual circular collimator. These were
7 cm high, 4.1 cm wide and 2.1 cm thick. Each one could be independently
translated and rotated about the mid-line under microprocessor control. Each pair
lay in a plane and anti-collision software ensured that no two collimators in a
plane ever collided, although two collimators not in the same plane could overlap
in the beam's-eye-view. Sufficient range of movement was engineered so that, in
the limit,the field could be either circular (all four vanes backed off beyond the
diameter of the circle), have four straight sides (all four vanes within the circle),or
one to three straight with three to one circular sides.
These eight degrees of movement (four translations; four rotations) enabled the
field to be dynamically shaped to the beam's-eye-view of the target at different
positions within any one arc and for each arc, thus protecting normal non-target
brain tissue. The concept is analogous to the idea behind the (large-field) multileaf
collimator (chapter 5 ) .
McGinley et a1 (1992) have made a similar,but not identical,development. The
group at the Emory Clinic have added the possibility of incorporating blocks
into the cylindrical collimator. These blocks are set by hand,rather than under
computer control.

Figure 3.21. Perspective view of the 'dynamic collimator' for stereotactic

radiosurgery with the field shaped to the beam's-eye-view of the target at
each arc position. The usual circular collimator is supplemented with four
independently positioned block collimators to define an eight-sidedjield (see
text). (FromLeavitt et a1 ( I 991).) (Reprintedwith permissionfrom Pergamon
Press Ltd, Ogord, UK.)
3.2.3. Stereotactic multiple-arc radiotherapy with a CT scanner

Iwamoto et a1 (1990) propose to use the CT scanner itself to perform multiple-
arc radiotherapy. By adding a secondary collimator to the CT scanner to define a
pencil beam, the machine's scanning movements naturally provided for rotation-
arc therapy. By tilting the gantry by up to f20" from the vertical plane, non-
coplanar arcs were possible. The advantageof this is that the need for a stereotactic
frame, or precise correlation between the geometries of imaging and therapy,
disappears. Images may be taken during therapy, confirming that the tumour is
being irradiated correctly. It tumed out that a CT image could be made with the
radiation which leaked into the full fan through the collimator, showing the target
volume as a black circle with a white halo and some detail of the rest of the slice.
Iwamoto et a1 (1990) compared the dose profile delivered this way with that from
two gamma knives in Stockholm and UCLA and noted little difference when nine
arcs were used. A typical dose of 2 Gy can be given in 20 rotations of 4 s duration
with the tube operating at 170 mA without overheating problems. Perhaps this
implies that the machine is more a rival for SMART than for the gamma knife,
since to deliver say 10 Gy for radiosurgery would require nearly an hour with
time for tube cooling. Account has to be made of the different RBE (1.5) of 140
kVp radiation.
3.2.4. Comparison of different radiation types and geometries

Which radiation type (charged particles or photons) and which geometry are best

Radiosurgery and stereotactic radiotherapy 16 1
for stereotactic radiosurgery? This is clearly an important question, given that the
technologies are expensive and that each has developed in its own separate way
at different treatment centres. Phillips er a1 (1990) studied this question. They
compared charged-particle irradiations from protons, carbon, helium and neon
ions with photon irradiations in the ‘gamma-knife’ geometry and with photon
irradiation from multiple arcs on a linear accelerator.
Dose distributions were developed for irradiating spherical tumours of
diameters 1 , 2 , 3 , 4 , 5 cm and for a range of tumours from clinical examples. For
the charged-particle irradiations the regular treatment geometries for these were
adopted. An optimal treatment modality is that which delivers a uniform dose
throughout the target volume whilst minimizing the dose to surrounding tissue,
and so the problem was analysed by computing the dose-volume histograms in
the target volume, in the ‘normal brain’ (being all the brain minus the tumour)
and for selected annular volumes surrounding the tumour. Additionally, the
figure of merit-the ‘localization factor’-was computed, being the fraction of
the radiation energy deposited in the patient which was delivered to the target
volume. The higher the localization factor, the better the treatment.
For very small tumours (of diameter 2 cm or less) virtually no differences
were observed between photon, proton and carbon irradiations. For medium size
(diameters 2-3 cm) and large size tumours (diameter 2 3 cm) photon irradiations
were increasingly inferior to charged-particleirradiations. For a tumour of volume
56 cm3 the localization factor for photons (0.20) was just over half that (0.38)
for protons, and for such large tumours the dose-volume histograms showed
considerably more normal-tissue (e.g. brainstem) dose with photons than with
protons. Only small differences were noted between the different charged-particle
modalities for stereotactic radiosurgery, these differences being greater for the
tumours with largest volumes. The biological effects of the major difference
between photons and protons has yet to be worked out, and Phillips et a1 (1990)
conclude that the efficacy of one type of radiation or irradiation scheme over
another is not yet clinically established.
Graham et a1 (1991a,b) used the Boston 3D planning system, JCPLAN, to assess
the relative merits of using 1,2,3,4,5, and 9 arc stereotactic irradiationscompared
with static coplanar three-field and static non-coplanar six-field techniques. The
dose distributions were characterized and ranked in terms of the integral d o s e
volume histogram in the normal brain lying in an annulus of thickness 2 cm
around spherical target volumes of size 10, 20, 40, and 55 mm diameter. It
had already been established that the DVH for these annulae were identical to
the whole-brain DVH, except at low doses, and the use of the annulus speeded
the computations. These target volumes matched the diameters of available
collimators. The different irradiation geometries were also ranked in terms of
the volume receiving 50% of the isocentre dose. (This value was chosen since the
brain may tolerate 10 Gy in a single fraction when the target dose-for a cerebral
AVM-may be 20 Gy.) Graham et a1 (1991a) concluded that (i) the distributions
with multiple arcs were superior to the three-static-fieldarrangements for all target

sizes, (ii) there was no increased sparing of normal brain tissue above the 50%
isodose with more than three arcs of rotation, and (iii) for the larger tumours, two
arcs were marginally better than three and the six-field static method also gave a
very similar distribution of dose to multiple-arc irradiations.
Serago et af (1992) also studied the effect of multiple arcs on the dose-volume
histogram for the normal brain surrounding spherical targets of diameter 1 , 2 and
4 cm. They carefully quantitated the volume of normal brain inside each isodose
contour as a function of number of arcs for each target size and concluded that
multiple arcs beyond four provided no significant improvement. There was a
marginal reduction in the volume of tissue enclosed by the 10% isodose contour
for 2 and 4 cm diameter targets, but this is clinically insignificant. Serago et af
(1992) concluded there may be a small safety advantage using multiple arcs, in
that the dose per arc is less and hence failure of the gantry to rotate would be less
serious. However, this event is unlikely to occur if proper safety precautions are
taken. Schell et a1 (1991) characterized dose distributions using the dose-volume
histogram, showing that, although differences were observed between 4, 5 and
10-arc therapy, these were insignificant clinically.
Chierego et a1 (1988) decided, on the basis of measuring, with film, the dose
distribution on selected planes (not volumes), that the best energy for use in
stereotactic radiotherapy was between 4-6 MV and that between 9 and 17 arcs
were generally needed. This conflict with the conclusions of Graham et al(1991a)
and Serago et al (1992) may arise because conclusions were made from planar
rather than volumetric dose measurements. Chierego et a1 (1988) measured the
absorbed dose to critical organs when a lesion sited in a deep medial cerebral plane
<
was irradiated and found the doses were negligible (all 0.2% of the isocentre
dose.)
Pike et af (1990) showed that the dose distributions obtained with multiple-arc
therapy were similar to those from the gamma knife. It was also demonstrated that
the distributions were virtually independent of photon energy in the range 4-25
MV, concurring with results by Mazal et af (1989).
In principle, the spherical sector which may be covered by a multiple-arc
technique can be larger than for the gamma knife. For example, Colombo et a f
(1985a) created spherical sectors of 110" x 160" compared with the 60" x 160"
of the gamma knife. How important this might be must be quantified for each
situation. Another potential advantage of using a linear accelerator is the ability
to use fields of larger area than provided by the collimators of the gamma knife.
Colombo et a1 (1990) introduced a modification to multiple-arc radiosurgery
enabling non-spherical dose distributions to be obtained. The radiation beam was
directed to the isocentre of the gantry and couch and, for a number of gantry
angles @, the couch was dynamically rotated through an angle a symmetrically
placed about the long axis of the patient (figure 3.22). By also arranging for
couch translation, cigar-shaped isodose distributions were obtained, more suitable
when the target volume was non-spherical. Casentini et a1 (1990) used the same
apparatus for arcing in the angle @ at a set of fixed angles a.

Stereotactic interstitial implant therapy 163
Figure 3.22. Stereotactic conical irradiation with a linear accelerator

arranged so that multiple-arc therapy would lead to a non-spherical dose
distribution. The gantry was set at a number of fixed angles 9, and the
patient couch was moved through an angle a. B y also arranging for couch
translation, elongated dose distributions were obtained. (From Colombo et
a1 ( I 990), permission of S Karger AG, Basel, Switzerland.)
3.3. STEREOTACTIC INTERSTITIAL IMPLANT THERAPY
For many tumour sites the high-dose volume can be conformed to the target
by interstitial radiotherapy; a brain tumour is one such candidate. This greatly
reduces the dose to non-target tissue, since extemal rays do not pass through this
tissue en route to the target. Stereotacticprocedures are again needed for accuracy.
Welsh and Chapman (1990) described procedures exemplifying the method. The
physical basis is as follows.
A stereotactic frame is attached to the head of the patient non-invasively. The
frame is also attached to the couch of a CT scanner and all procedures take place
in the CT scanner suite. To the frame is attached a template comprising a matrix
of holes spaced at 1.5 mm intervals. These holes are candidate drilling sites for
catheterizing the brain. CT images are taken with the frame and template in place;
the template appears suspended above the skull and each aperture appears as an
air-filled column on the template. From these images, a suitable set of apertures
can be selected for approaching the target. The skull is drilled and the catheters are
introduced to a depth of 1 cm beyond the contrast-enhancedborder of the tumour.
When all catheters are in place, a repeat set of CT scans is made to establish
that the catheters are all parallel, correctly placed and that there has been no
intraprocedural haemmorhage. Finally, the catheters are secured externally. The
whole procedure takes place under general anaesthetic and takes about 2 hours. A
widely used radioisotope is 12’1, in the form of seeds. This isotope has a half-life
of 60 days and emits x-rays of 28 keV (among other less abundant radiations).

Figure 3.23. The BRW stereotactic system in place, with the template
guiding the placement of several catheters. These could be catheters
holding radioactive seeds (1251,
'921r)or ferromagnetic seeds for inductive
hyperthermia. (From Lulu et a1 ( I 990).) (Reprinted with permission from
Pergamon Press Ltd, Ogord, UK.)
The catheters with the seeds are removed after the required dose has been given,
and the patient is CT scanned once again. Lulu et a1 (1990) and Marchosky et a1
(1989) have described a similar CT-based stereotactic procedure for guiding the
placement of catheters and the associated treatment planning. Figure 3.23 shows
the BRW stereotactic system in place with a template and several catheters.
Hawliczek et af (1991) have presented a technique for implanting seeds
permanently in the brain. Planning methods have been evolved to display the
isodoses from candidate seed positions superposed on anatomical images (Weaver
et al 1990) and then to calculate through coordinate transformations the entry
trajectories in the frame coordinates for catheters. Ling er a1 (1985) and Schell
er af (1987) have provided two-dimensional dose matrices for '"I dosimetry
calculations.
3.4. SUMMARY
Stereotactic radiotherapy and radiosurgery were proposed over four decades ago.
Although the earliest treatments were made with x-rays, the specialist gamma
knife was for many years the only reliable means for treatment. However,
the achievement of stereotactic conformal radiotherapy and radiosurgery using
a linear accelerator with special fixation devices has recently given the subject
an enormous boost, and today there is widespread interest in the technique. In
tum this has led to characterizing the effectiveness of the method in terms of
well understood outcomes such as the dose-volume histogram of dose to target
and surrounding tissue and plots of the dose profiles in selected planes. These

References 165
have led many workers to draw conclusions on the optimum number of arcs as a
function of target size. Special techniques are being developed to generate non-
spherical treatment volumes, where required. Image-guided stereotactic implants
are also performed. In many circumstances these methods can be as satisfactory
as charged-particle irradiation.
REFERENCES
Amdt J 1989 Use of the linear accelerator for neurosurgery Neurosurgery 24 (4)
64 1
Barrow D L, Bakay R A E, Crocker I, McGinley P and Tindall G T 1990
Stereotactic radiosurgery J. Med. Assoc. of Georgia 79 667-676
Bergstrom M, Boethius J, Eriksson L, Greitz T, Ribbe T and Widin L 1981 Head
fixation device for reproducible position alignment in transmission CT and
positron emission tomography J. Comp. Assist. Tomog. 5 (1) 136-141
Betti 0 0 and Derechinsky Y E 1982 Irradiations stdreotaxiques multifaisceaux
Neurochirurgie 28 55-56
Betti 0 0, Munari C and Rossler R 1989 Stereotactic radiosurgery with the linear
accelerator: treatment of arteriovenous malformations Neurosurgery 24 3 11-
32 1
Bova F J and Friedman W A 1991 Stereotactic angiography: an inadequate
database for radiosurgery? Int. J Rad. Oncol. Bioi. Phys. 20 891-895
Brown R A 1979 A stereotactic head frame for use with CT body scanners
Investigative Neurology 14 30G304
van Buren J M, Landy H J, Houdek P V and Ginsberg M S 1986 CT-
directed stereotactic fractionated rotational radiotherapy by linear accelerator
Neurosurgery 19 (1) 149
Casentini L, Colombo F, Pozza F and Benedetti A 1990 Combined radiosurgery
and extemal radiotherapy of intracranial germinomas Surgical Neurology 34
(2) 79-86
Chierego G, Marchetti C, Avanzo R C, Pozza F and Colombo F 1988 Dosimetric
considerations on multiple arc stereotaxic radiotherapy Radiother. Oncol. 12
141-152
Clarke R H 1914 Stereotactic instrument UK Patent no 11711
Coffey R J, Flickinger J C, Bisonette D J and Lunsford L D 1991 Radiosurgery for
solitary brain metastases using the cobalt-60 gamma unit: methods and results
in 24 patients Int. JRad. Oncol. B i d . Phys. 20 1287-1295
Coffey R J, Lunsford L D, Bisonette D and Flickinger J C 1990 Stereotactic
gamma radiosurgery for intracranial vascular malformations and tumours:
report of the initial North American Experience in 33 1 patients Stereotactic and
functional neurosurgery ed C Ohye, P L Gildenberg and P 0 Franklin (Basel:
Karger) pp 535-540
Colombo F, Benedetti A, Casentini L, Zanusso M and Pozza F 1987 Linear

accelerator radiosurgery of arteriovenous malformations Appl. Neurophysiol.

50 257-261
Colombo F, Benedetti A, Pozza F, Avanzo R C, Marchetti C, Chierego G
and Zanardo A 1985a Extemal stereotactic irradiation by linear accelerator
Neurosurgery 16 154-160
Colombo F, Benedetti A, Pozza F, Marchetti C, Chierego G and Zanardo A
1990 Linear accelerator radiosurgery of three-dimensional irregular targets
Stereotactic and functional neurosurgery ed C Ohye, P L Gildenberg and P
0 Franklin (Basel: Karger) pp 541-546
Colombo F, Benedetti A, Pozza F, Zanardo A, Avanzo R C, Chierego G and
Marchetti C 1985b Stereotactic radiosurgery utilising a linear accelerator
Applied Neurophysiol. 48 133-145
Croft M J 1990 Stereotactic radiosurgery of arteriovenous malformations
Radiologic Technol. 61 375-379
Delannes M, Daly N, Bonnet J, Sabatier J and Trdmoulet M 1990 Le sttdo-
adapteur de Laitinen Neurochirurgie 36 167-175
Drzymala R E 1991 Quality assurance for linac-based stereotactic radiosurgery
Quality assurance in radiotherapy physics ed G Starkschall and J Horton
(Madison, WI: Medical Physics Publishing) pp 121-129
Engenhart R, Kimmig B N, Hover K H, Wowra B, Sturm V, Kaick G and
Wannenmacher M 1990 Stereotactic single high dose radiation therapy of
benign intracranial meningiomas Int. J . Rad. Oncol. Bid. Phys. 19 1021-1026
Flickinger J C, Lunsford L D, Wu A and Kalend A 1991 Predicted dose- volume
isoeffect curves for stereotactic radiosurgery with the 6oCo gamma unit Acta
Oncol. 30 363-367
Flickinger J C, Lunsford L D, Wu A, Maitz A H and Kalend A M 1990a Treatment
planning for gamma knife radiosurgery with multiple isocentres Int. J . Rad.
Flickinger J C, Maitz A, Kalend A, Lunsford L D and Wu A 1990b Treatment
volume shaping with selective beam blocking using the Leksell Gamma Unit
Flickinger J C and Steiner L 1990 Radiosurgery and the double logistic product
formula Radiother. Oncol. 17 229-237
Fodstad H, Hariz M and Ljunggren B 1990 History of Clarke’s stereotactic
instrument Stereotactic and functional neurosurgery ed C Ohye, P L
Gildenberg and P 0 Franklin (Basel: Karger) pp 242
Fox P T, Perlmutter J S and Raichle M E 1985 A stereotactic method of anatomical
localization for positron emission tomography J . Comp. Assist. Tomog. 9 (1)
141-153
Friedman W A and Bova F J 1989 The University of Florida radiosurgery system
Surg. Neurol. 32 334-342
Galloway R L and Maciunas R J 1990 Stereotactic neurosurgery Critical Reviews
in Biomedical Engineering 18 (3) 181-205
Gehring M A, Mackie T R, Mehta M P, Kubsad S S , Sanders C A and Reckwerdt

References 167
P S 1991 A 3D treatment planning system for stereotactic radiosurgery Proc.

1st biennial ESTRO meeting on physics in clinical radiotherapy (Budapest,
1991) p 29
Gill S S 1987 UK Patent application no 8728150
Gill S S , Thomas D G T, Warrington A P and Brada M 1991 Relocatable frame for
stereotactic extemal beam radiotherapy Int. J . Rad. One B i d . Phys. 20 599-603
Goodman M L 1990 Gamma knife radiosurgery: current status and review
Southern Medical J . 83 ( 5 ) 551-554
Graham J D, Nahum A E and Brada M 1991a A comparison of techniques for
stereotactic radiotherapy by linear accelerator based on 3-dimensional dose
distributions Radiother. Oncol. 22 29-35
Graham J D, Warrington A P, Gill S S and Brada M 1991b A non-invasive
relocatable stereotactic frame for fractionated radiotherapy and multiple
imaging Radiother. Oncol. 21 60-62
Greitz T, Bergstrom M, Boethius J, Kingsley D and Ribbe T 1980 Head
fixation system for integration of radiodiagnostic and therapeutic procedures
Neuroradiology 19 1-6
Hartmann G H, Schlegel W, Sturm V, Kober B, Pastyr 0 and Lorenz W J 1985
Cerebral radiation surgery using moving field irradiation at a linear accelerator
facility Int. J . Rad. Oncol. B i d . Phys. 11 1185-1192
Hartmann G H, Treuer H, Boesecke R, Bauer B, Sturm V and Lorenz W J
1987 'Problems of dose calculation applying stereotactic convergent beam
irradiations at head neck region Computer Assisted Radiology ed H U Lemke,
M L Rhodes, C C Jaffee and R Felix (Berlin: Springer) pp 328-334
Hanaway J, Scott W R and Strother C M 1980 Atlas of the human brain and the
orbit for computed tomography (St Louis, Missouri: Warren H Green Inc)
Hariz M I, Henriksson R, Lijfroth P 0, Laitinen L V and Saterborg N E 1990 A
non-invasive method for fractionated stereotactic irradiation of brain tumours
with linear accelerator Radiother. Oncol. 17 57-72
Hariz M I, Laitinen L V Lofroth P 0 and Saterborg N E 1988 A non-invasive
adapter in the stereotactic irradiation of brain tumours with linear accelerator
Acta Neurochirur. 91 (3/4) 162
Hawliczek R, Neubauer J, Schmidt W F 0, Grunert P and Coia L R 1991 A new
device for interstitial '2510dineseed implantation Znt. J. Rad. Oncol. B i d . Phys.
20 621-625
Heifetz M D, Wexler M and Thompson R 1984 Single-beam radiotherapy knife
J . Neurosurgery 60 8 14-8 18
Heifetz M D, Whiting J, Bemstein H, Wexler M, Rosemark P and Thomson R W
1989 Stereotactic radiosurgery for fractionated radiation: a proposal applicable
to linear accelerator and proton beam programs Stereotact. Func. Neurosurgery
53 167-177
Heilbrun M P, Roberts T S , Apuzzo M L J, Wells Jr T H and Sabshin J K 1983
Preliminary experience with the Brown-Roberts-Wells (BRW) computerized
tomography stereotaxic guidance system J . Neurosurgery 59 2 17-222

Henri C, Collins L, Peters T M, Evans A C and Marrett S 1989 Three-dimensional

interactive display of medical images for stereotactic neurosurgery planning
Proc. SPZE 1092 (Medical Imaging 3: Image Processing) 67-74
Herbert M H and Froder M 1990 Digital tumour fluoroscopy (DTF a-) new direct
imaging system in the therapy planning for brain tumours Znt. J. Rad. Oncol.
Biol.Phys. 18 221-231
Horsley V and Clarke R H 1908 The structure and functions of the cerebellum
examined by a new method Brain 31 45-124
Iwamoto K S , Norman A, Kagan A R, Wollin M, Olch A, Bellotti J, Ingram M
and Skillen R G 1990 The CT scanner as a therapy machine Radiother. Oncol.
19 337-343
Kingsley D P E, Bergstrom M and Berggren B M 1980 A critical evaluation of
two methods of head fixation Neuroradiology 19 7-12
Kondziolka D and Lunsford L D 1991 Stereotactic radiosurgery for squamous cell
carcinoma of the nasopharynx Laryngoscope 101 5 19-522
Kondziolka D, Lunsford L D, Coffey K J, Bisonette D J and Flickinger J C 1990
Stereotactic radiosurgery of angiographically occult vascular malformations:
indications and preliminary experience Neurosurgery 27 (6) 892-900
Kubsad S S , Mackie T R, Gehring M A, Misisco D J, Paliwal B R, Mehta M P
and Kinsella T J 1990 Montecarlo and convolution dosimetry for stereotactic
radiosurgery Int. J. Rad. Oncol. Biol. Phys. 19 1027-1035
Larsson B, Leksell L, Rexed B, Sourander P,Mair W and Anderson B 1958 The
high-energy proton beam as a neurosurgical tool Nature 182 1222-1223
Leavitt D D, Gibbs FA, Heilbrun M P, Moeller J H and Takach G A 1991 Dynamic
field shaping to optimise stereotactic radiosurgery Znt. J . Rad. Oncol. Biol.
Phys. 21 1247-1255
Leksell D G 1987 Stereotactic radiosurgery Neurological Research 9 60-68
Leksell L 1949 A stereotaxic apparatus for intracerebral surgery Acta Chirurg.
Scand. 99 229-233
-1951 The stereotaxic method and radiosurgery of the brain Acta Chirurg.
Scand. 102 316-319
-1971 Stereotaxis and radiosurgery (Springfield, IL: Charles C Thomas)
Ling C, Schell M, Yorke E, Palos B and Kubiatowicz D 1985 Two-dimensional
dose distribution of Iz5Iseeds Med. Phys. 12 652-655
Loeffler J S , Rossitch E Jr, Siddon R, Moore M R, Rockoff M A, Alexander 111E
1990a Role of stereotactic radiosurgery with a linear accelerator in treatment
of intracranial arteriovenous malformations and tumours in children Pediatrics
85 774-782
Loeffler J S , Siddon R L, Wen P Y, Nedzi L A and Alexander I11 E 1990b
Stereotactic radiosurgery of the brain using a standard linear accelerator: a
study of early and late effects Radiother. Oncol. 17 311-321
Lulu B A, Lutz W, Stea B and Cetas T C 1990 Treatment planning of template-
guided stereotaxic brain implants Znt. J. Rad. Oncol. Biol. Phys. 18 95 1-955
Lunsford L D, Coffey R J, Cojocaru T and Leksell D 1990a Image-guided

References 169
stereotactic surgery: a 10 year evolutionary experience Stereotactic and

functional neurosurgery ed C Ohye, P L Gildenberg and P 0 Franklin (Basel:
Karger) pp 375-387
Lunsford L D, Flickinger J and Coffey R J 1990b Stereotactic gamma knife
radiosurgery Arch. Neurol. 47 169-175
Lutz W, Winston K R and Maleki N 1988 A system for stereotactic radiosurgery
using a linear accelerator Int. J . Rad. Oncol. Biol. Phys. 14 373-381
Luxton G, Jozsef G and Astrahan M A 1991 Algorithm for dosimetry of multiarc
linear-accelerator stereotactic radiosurgery Med. Phys. 18 1211-1221
Maitz A H, Lunsford L D, Wu A, Lindner G and Flickinger J C 1990 Shielding
requirements on-site loading and acceptance testing of the Leksell Gamma
Knife Int. J. Rad. Oncol. Biol. Phys. 18 469476
Marchosky J A, Moran C, Feamot N E, Welsh D M and Zumwalt C B 1989 A
stereotactic system for volumetric implantation of the brain Proc. 8th Ann.
Meeting North American Hyperthermia Group 1 47
Matsui T and Hirano A 1978 An atlas of the human brain for computerized
tomography (Tokyo: Igaku-Shoin)
Mazal D A, Rosenwald J C, Gaboriaud G and Porcher J B 1989 Implication of
basic physical phenomena on the dose distribution in stereotactic x-ray extemal
irradiation Proc. 17th Int. Congress of Radiology p 217
Mazziotta J C, Phelps M E, Meadors A K, Ricci A, Winter J and Bentson J R 1982
Anatomical localisation schemes for use in positron computed tomography
using a specially designed headholder J. Comp. Assist. Tomog. 6 (4) 848-853
Meltzer C C, Bryan R N, Holcomb H H, Kimball A W, Mayberg H S , Sadzot B,
Leal J P, Wagner H N and Frost J J 1990 Anatomical localisation for PET using
MR imaging J . Comp. Assist. Tomog. 14 (3) 418426
McGinley P H, Butker E K, Crocker I R and Aiken R 1992 An adjustable
collimator for stereotactic radiosurgery Phys. Med. Biol. 37 4 13-41 9
McGinley P H, Butker E K, Crocker I R and Landry J C 1990 A patient rotator
for stereotactic radiosurgery Phys. Med. Biol. 35 649-657
Miura S , Kanno I, Iida H, Marakami M, Takahashi K, Sasaki H, Inugami A,
Shishido F, Ogawa T and Uemura K 1988 Anatomical adjustments in brain
positron emission tomography using CT images J . Comp. Assist. Tomog. 12 (2)
363-367
Pastyr 0, Hartmann G H, Schlegel W, Schabbert S , Treuer H, Lorenz W J and
Sturm V 1989 Stereotactically guided convergent beam irradiation with a linear
accelerator: localisation technique Acta Neurologica 99 6 1-64
Patil A A 1985 Isocentric placement of the target in the linear accelerator using
CT stereotaxis Acta Neurochirurgica 78 168-1 69
Patil A A 1989 Radiosurgery with the linear accelerator Neurosurgery 25 143
Phillips M H, Frankel K A, Lyman J T, Fabrikant J I and Levy R P
1990 Comparison of different radiation types and irradiation geometries in
stereotactic radiosurgery Int. J. Rad. Oncol. Biol. Phys. 18 21 1-220
Pike B, Peters T M, Podgorsak E, Pla C, Olivier A and de LotbiniBre A 1987

Stereotactic external beam calculations for radiosurgical treatment of brain

lesions Appl. Neurophysiol. 50 269-273
Pike G B, Podgorsak E B, Peters T M, Pla C and Olivier A 1990 Three-dimensional
isodose distributions in stereotactic radiosurgery Stereotactic and functional
neurosurgery ed C Ohye, P L Gildenberg and P 0 Franklin (Basel: Karger) pp
5 19-524
Podgorsak E B, Olivier A, Pla M, Hazel J, de Lotbinikre A and Pike B 1987
Physical aspects of dynamic stereotactic radiosurgery Appl. Neurophys. 50
263-268
Podgorsak E B, Olivier A, Pla M, Levebvre P Y and Hazel J 1988 Dynamic
stereotactic radiosurgery Int. J . Rad. Oncol. B i d . Phys. 14 115-126
Podgorsak E B, Pike B, Olivier A, Pla M and Souhami L 1989 Radiosurgery with
high energy photon beams: A comparison among techniques Int. J. Rad. Oncol.
B i d . Phys. 16 857-865
Saunders W M, Winston K R, Siddon R L, Svensson G H, Kijewski P K, Rice R K,
Hansen J L and Barth N H 1988 Radiosurgery for arteriovenous malformations
of the brain using a standard linear accelerator: Rationale and technique Int. J .
Schell M C, Ling C, Gromadzki Z and Working K 1987 Dose distributions of
model 6702 1-125 seeds in water Int. J. Rad. Oncol. Biol. Phys. 13 795-799
Schell M C , Smith V, Larson D A, Wu A and Flickinger J C 1991 Evaluation
of radiosurgery techniques with cumulative dose volume histograms in linac-
based stereotactic external beam irradiation Int. J . Rad. Oncol. Biol. Phys. 20
1325-1330
Schwade J G, Houdek P V, Landy H J, Bujnoski J L, Lewin A A, Abitol A
A, Serago C F and Pisciotta V J 1990 Small-field stereotactic external-beam
radiation therapy of intracranial lesions: fractionated treatment with a fixed halo
device Radiology 176 563-565
Serago C F, Houdek P V, Bauer-Kirpes B, Lewin A A, Abitol A A, Gonzales-Arias
S , Marcial-Vega V A and Schwade J G 1992 Stereotactic radiosurgery: dose-
volume analysis of linear accelerator techniques Med. Phys. 19 (1) 181-1 85
Serago C F, Lewin A A, Houdek P V, Gonzales-Arias S , Hartmann G H, Abitol A
A and Schwade J G 1991 Stereotactic target point verification of an x-ray and
CT localiser Int. J . Rad. Oncol. Biol. Phys. 20 517-523
Souhami L, Olivier A, Podgorsak E B, Hazel J, Pla M and Tampieri D 1990a
Dynamic stereotaxic radiosurgery in arteriovenous malformation-preliminary
treatment results Cancer 66 (1) 15-20
Souhami L, Olivier A, Podgorsak E B, Pla M and Pike G B 1990bRadiosurgery of
cerebral arteriovenous-malformations with the dynamic stereotaxic irradiation
Spiegel E A, Wycis H T, Marks M and Lee A J 1947 Stereotaxic apparatus for
operations on human brain Science 106 349-350
Talairach J and Szikla G 1967 Atlas d’anatomie stkrkotaxique du tklenckphale:
etudes anatomio-radiologiques (Paris; Masson et Cie)

References 171
Talairach J and Toumoux P 1988 Co-planar stereotaxic atlas of the human brain
(Stuttgart: Georg Thieme)
Thomas D G T, Gill S S, Wilson C B, Darling J L and Parkins C S 1990 Use of
relocatable stereotactic frame to integrate positron emission tomography and
computed tomography images: application in human malignant brain tumours
Stereotactic and functional neurosurgery ed C Ohye, P L Gildenberg and P 0
Franklin (Basel: Karger) pp 388-392
Thomson E S 1990 Stereotactic brain radiotherapy PhD thesis University of
London
Thomson E S , Gill S S and Doughty D 1990 Stereotactic multiple arc radiotherapy
Brit. J . Radiol. 63 745-75 1
To S Y C, Lufkin R B, Rand R, Robinson J D and Hanafee W 1990 Volume growth
rate of acoustic neuromas on MRI post-stereotactic radiosurgery Comp. Med.
Imag. Graph. 14 53-59
Tsai J S , Buck B A, Svensson G K, Alexander 111 E, Cheng C-W, Mannarino E
G and Loeffler J S 1991 Quality assurance in stereotactic radiosurgery using a
standard linear accelerator Znt. J . Rad. Oncol. B i d . Phys. 21 737-748
Walton L, Bomford C K and Ramsden D 1987 The Sheffield stereotactic
radiosurgery unit: physical characteristics and principles of operation Brit. J .
Radiol. 60 897-906
Weaver K, Smith V, Lewis J D, Lulu B, Bamett C M, Leibel S A, Gutin P, Larson
D and Phillips T 1990 A CT-based computerized treatment planning system for
lZ5Istereotactic brain implants Znt. J . Rad. Oncol. B i d . Phys. 18 445-454
Welsh D and Chapman D 1990 A stereotactic technique for volumetric interstitial
implantation in the brain J. Neuroscience Nursing 22 (4) 245-249
Wilson M W and Mountz J M 1989 A reference system for neuroanatomical
localisation on functional reconstructed cerebral images J . Comp. Assist.
Tomog. 13 (1) 174-178
Winston K R and Lutz W 1988 Linear accelerator as a neurosurgical tool for
stereotactic radiosurgery Neurosurgery 22 (3) 454464
Wu A, Lindner G, Maitz A H, Kalend A M, Lunsford L D, Flickinger J C and
Bloomer W D 1990 Physics of gamma knife approach on convergent beams in
stereotactic radiosurgery Int. J . Rad. Oncol. B i d . Phys. 18 941-949
Wu A, Maitz A H and Kalend A M 1991 The role of the gamma knife in
radiosurgery-physicist’s perspective Proc. 1st biennial ESTRO meeting on
Zhang J, Levesque M F, Wilson C L, Harper R M, Engel J, Lufkin R and Behnke
E J 1990 Multimodality imaging of brain structures for stereotactic surgery
Radiology 175 4 3 5 4 1

CHAPTER 4
THE PHYSICS OF PROTON

RADIOTHERAPY
4.1. INTRODUCTION: ELEMENTARY PHYSICS OF PROTON BEAMS
All radiotherapy aims to achieve a high therapeutic ratio, defined as the ratio of
the probability of tumour control to the probability of normal tissue complication.
Ways to do this with x-rays rely on arranging that the dose to organs-at-risk
is significantly lower than the dose to the target volume by suitable multiport,
rotational, dynamic or other techniques. These include those involving shaping
the geometric field to the beam’s-eye-view of the target, whose ultimate aim
is dose conformation to the target volume. However, all radiotherapy with x-
rays is frustrated by the physics of the interaction of radiation with matter. This
determines that when a single beam irradiates a target volume, the dose on
the proximal side of the target volume is always higher than that in the target
volume and the dose on the distal side, although lower, is certainly not zero.
For this reason, even combinations of beams do not always give acceptable dose
distributions.
Radiotherapy with charged particles is fundamentally different and electrons,
protons, heavy negative ions and helium ions have all been used. When a proton
beam interacts with tissue, most of its energy is lost by collision with atomic
orbital electrons. Since the proton is some 1835 times more massive than the
electron, it does not significantly deviate from a straight-line path as it interacts,
slows down and eventually stops. The proton deposits energy with an energy loss
inversely proportional to the square of its speed. The typical distribution of dose
with depth (figure 4.1) exhibits a strong Bragg peak towards the very end of the
range, beyond which the dose very rapidly falls to zero and on the proximal side
of which, the dose is only some 20% of the peak dose. Figure 4.2 emphasizes
inherent differences between photon and proton irradiation. The pattern of dose
deposition was first observed by Bragg and Kleeman (1904). In principle this is
an ideal arrangement for radiotherapy if a target can be arranged to coincide with
the Bragg peak. To give a quantitative feel for this, the range of 160 MeV protons
in tissue is about 16 cm and the width of the Bragg peak at the 80% level is only
172
Introduction: elementary physics of proton beams 173
Modulated Proton Beam
100 -
80 -
d
8 60-
n
?!
.
-‘2 40-
2
20 -
Unmodulated Proton Beam
0- I I 1 I
0 4 8 12 16
Depth in Tissue, cm
Figure 4.1. Depth-dose curve for an unmodulated and a modulated 160

MeV proton beam, compared with that for a 10 MV x-ray beam. The x-ray
beam falls exponentially after the initial build-up, whereas the maximum in
the proton curve rises slowly to the Bragg peak where the protons stop. I f
the proton beam is modulated, the Bragg peak spreads out but the dose on
the proximal side of the Bragg plateau is then larger thanfor monoenergetic
protons. The advantage of the proton beam is the complete absence of dose
beyond the Bragg peak and the avoidance of surface dose. (From Verhey and
Munzenrider ( I 982).)(Reproduced, with permission,from the Annual Review
of Biophysics and Bioengineering; Vol I 1 0 1 9 8 2 by Annual Reviews Inc.)
Figure 4.2. Illustrates the fundamental advantage of proton therapy. In the

upper figure a target is irradiated to dose D2 by photons. The dose proximal
to the target is higher than the target dose and the distal dose is non-zero. In
the lower figure the same target is irradiated to dose D2 by protons. The dose
proximal to the target is now lower than the target dose and the distal dose
is absolutely zero when the beam is shaped to the distal surface of the target
volume. Combinations of photon beams can improve on the upper situation
of course, but the single proton beam possesses this fundamental advantage.
7 mm (Verhey and Munzenrider 1982).

About 1-296 of the energy loss arises from collisions with the atomic nuclei
during which a much larger fraction of the proton’s energy is transferred to the

174 The physics of proton radiotherapy
medium. This causes a high linear-energy-transfer (LET) and in tum a higher

radiobiological effectiveness (RBE)than the electron-collision component which
is identical to that of x-rays (RBE=1). A commonly adopted value for the RBE of
a proton beam is about 1.1, determined from clinical studies with fractionation.
This value is however not certain and Raju (1980) provides a large table of the
biological results to that date. Values ranging between 0.6 and 1.4 have been
found. A commonly used unit is the ‘Cobalt Gray Equivalent (CGE)’, being the
dose in proton gray multiplied by 1.1, which gives the dose in cobalt gray (Cole
1990). Because the RBE for protons is similar to that of x-rays, ‘conventional’
x-ray clinical experience can be translated into proton-therapy management. So
protons present a challenging opportunity to improve conformal radiotherapy
because of essentially the better dose distributions which can be obtained (Kantor
et a1 1985) and the slightly enhanced RBE. Because dose distributions can be
more tightly conformed to the target volume, dose escalation can be practised.
At the Harvard Cyclotron Laboratory radiation doses of 70 CGE are regularly
prescribed. Koehler and Preston (1972) show many examples of comparative
dose distributions with x-rays and protons showing the advantages of the latter.
The advantages of proton irradiation over photons for stereotactic radiosurgery
are discussed in chapter 3 (section 3.2.4).
The stopping power of particles depends on the square of the ratio of their charge
to their velocity, so, once the stopping power of protons is established, the stopping
power of other particles follows from this law. For example, at the same velocity,
the values for helium ions (doubly charged alpha particles) are four times larger
but the values are the same for protons and deuterons, which have the same charge.
The range of protons in tissue depends on their energy (see Janni 1982, Bichsel
1972). A convenient formula to use is
R = 1 . 1 1 x lo-’ (E/9.29)’.8 (4.1)
where R is the proton range in tissue in cm and E is the proton energy in MeV
(Wilson 1946). Some specific values are in table 4.1 and figure 4.3 shows range
in water as a function of energy. The range of two different elementary particles
of the same velocity but different charge and mass is proportional to the mass
and inversely proportional to the square of the charge (Wilson 1946, Raju 1980).
Hence figure 4.3 can be used for deuterons if the energy axis and the range axis are
scaled by a factor of two. For example, a 200 MeV deuteron has a 16 cm range.
Figure 4.3 can be used for alpha particles if the energy axis is scaled by a factor
of four and the range axis stays the same. For example, a 400 MeV alpha particle
has a range of 8 cm.
Low-energy protons can be used for sites such as the treatment of occular
melanoma. High-energy protons are needed for deep sites. The Bragg peak of
monoenergetic protons is too narrow to be of use unmodified, Practical proton
beams for radiotherapy have range modifiers in the beam-line which stretch out the
Bragg peak. These are usually rotating propellers or paddles of variable thickness,

Table 4.1. The variation ofproton range in muscle

with proton energy (from Bonnett (1991)).
Energy (MeV) Range in ICRU Muscle (cm)

~
50 2.2
100 7.7
150 15.8
200 26.0
250 38.0
0 10 20 30
Range in water (cm)
Figure 4.3. The variation with proton energy of the proton range in water
(from Raju ( 1 980)).
made of plastic, which develop a spectral beam (see section 4.3.1). The so-called
‘Spread Out Bragg Peak’ (SOBP) can be arranged to straddle a width appropriate to
the target volume by constructing the appropriate beam modifier. Unfortunately,
spreading the Bragg peak also causes the dose on the proximal side of the target
volume to be significantly increased (see figure 4.1). In order to achieve a high
ratio between the dose to the target volume and the surrounding tissues, multiple-
proton fields can be utilized.
The large mass of the proton relative to the electron leads to a smaller range
straggling (2: 1%) than for the latter, a penumbra which is comparable with that
of x-radiation and much tighter than for electrons. Typically the 20% and 80%
isodose contours on the lateral edge are only 6 mm apart at 160 MeV. Penumbra
depends on the proton collimator and increases with depth. Coutrakon et a1
(1991a) have measured the penumbra (90%-10%) for the Loma Linda proton
facility as 6 mm at a depth of 3 cm in water and 15 mm at a depth of 25 cm.
Electron radiotherapy does not enjoy these advantages. Heavy-ion and negative-
pion therapy have a different biological effect and are not considered here.

4.1.1. Protons for clinical problems

Clinically, proton-beam radiotherapy is worth considering when the organs at risk
are extremely proximal to the target volume, circumstances when x-radiotherapy
is problematic because it is difficult to arrange to irradiate the target volume
without also irradiating the organs at risk. With x-radiation, even if the primary
beams do not irradiate the organs at risk, there is dose from scattered radiation.
Examples of such circumstances are choroidal melanomas and sarcomas abutting
critical central nervous structures.
For these problems, proton therapy has been considered justified without the
need to perform randomized trials comparing it with x-radiotherapy. Proton
therapy can treat occular melanoma without the need for enucleation with
disfigurement. Proton therapy is a strong candidate for treating chordoma,
soft-tissue sarcoma of the retroperitoneum, paranasal tissues and head and
neck, glioblastoma multiforme and meningioma (Cole 1990). Experience at
the Massachusetts General Hospital and Harvard Cyclotron Laboratory at these
clinical sites has led Suit et ai (1990) to write ‘Our assessment...,.with proton-
beam therapy over the period of 1974 to the present (1990) is that for the anatomic
sites tested there have been increases in the frequency of tumour control, lower
frequencies of treatment-related morbidity, and no increases in marginal failures.’
Since proton therapy is very expensive (and has its critics) such remarks are
valuable and the paper by Suit et a1 (1990) documents clinical experience in detail.
For many other clinical problems it is relatively easy to show that better dose
distributions can be obtained by proton therapy compared with x-radiotherapy
(figure 4.4) but the clinical benefit in terms of improved therapeutic ratio is less
obvious. In these cases it is necessary to arrange clinical randomized trials. Since
the number of proton facilities worldwide is quite small, the necessary clinical
data take time to accumulate (Antoine 1987). It is nearly 40 years since the first
treatment with protons and the worldwide number of patients treated stood at
(only) 10358 in Jan 1992. Indeed Goitein and Munzenrider (1984) argue that
such trials would have to be multicentric and international and various efforts
are underway to formulate a framework for such collaboration (Antoine 1991).
In particular, the PTCOG (Proton Therapy Cooperative Group) (Gottschalk 1987)
meets regularly. In a sense there is a chicken-and-egg situation where more proton
facilities are needed just to do the trials to determine whether proton irradiation is
worthwhile.
Raju (1980) gives three reasons why he feels that proton therapy has not really
‘taken off’ clinically:
1. proton therapy was first mooted in the 1940’s just as megavoltage photon
therapy was being considered for the first time,
2. high-energy nuclear physics was so well funded in the 1950’s and 1960’sthat
accelerators were in full-time use for nuclear physics and beam time could not be
spared for radiotherapy,
3. poor experience with neutron therapy clouded the issues.

ID
1.0 1.0
0' x
Figure 4.4. Comparative dose distributions for parallel-opposed proton

portals (left) and for a three-field 16 MV x-ray plan with lateral weighting
and wedging (right). The target volume is the hatched region. Notice how the
isodose contours relate to the location of the spinal chord. The proton plan
produces a distribution which irradiates normal structures much less thanfor
the x-ray plan. (From Verhey and Munzenrider (1982).)(Reproduced, with
permission, from the Annual Review of Biophysics and Bioengineering; Vol
I1 Q1982 by Annual Reviews Inc.)
Planning for proton therapy has many features similar to planning extemal-
beam x-radiotherapy. It must be image-driven with excellent delineation of
target volumes and organs at risk. Generally, CT images provide the necessary
data. Interestingly, because some facilities use stationary beam-lines in the
horizontal position and the patient must be correspondingly seated or standing for
treatment, two special-purpose CT scanners have been constructed which rotate
in the horizontal plane. These have been installed at the Harvard Cyclotron
Centre/Massachusetts General Hospital (HCC-MGH) and at the Lawrence
Berkeley Laboratory (LBL) (Goitein et a1 1982, Chen 1983). That the early
proton facilities have static beam-lines is reminiscent of a similar situation in the
early days of photon linear accelerators when the accelerator guide was large and
vacuum systems were a further limitation (Sharma et af 1991).
Some differences in proton therapy planning include the need to select optimum
entrance ports and to tailor beam-defining collimators from bronze blanks or
Cerrobend alloy. The patient is bolussed with material designed to create a
compensator which accounts for the different paths of the beam through the
patient, since tissue inhomogeneities, having electron densities different from
water, will modify the proton range. It has to be arranged that the combined
path length of bolus and different tissues between the beam port and the proximal
edge of the target volume is the minimum proton range in the spectral beam.
Compensators may be fabricated from lucite (Munzenrider et a1 1985) (see section
4.4.2). Similar considerations apply for radiotherapy with other heavy ions (Chen
et a1 1979).

t r c o f i 0 - 1
t DEPTH -
CHARGED -
PART I CLES
D E P ~ H-t
Figure 4.5. The effect of a 3 cm slab of bone on the depth-dose distribution

of a 6oCo beam compared with a proton beam. In the former the intensity of
rhe radiation is reduced; in the latter, the intensity is substantially unchanged
but the range is reduced. (From Goitein (1982).)
Tissue inhomogeneities present a different problem for proton therapy than for
x-ray therapy. In the latter it is the intensity of x-rays which is modulated by the
presence of an inhomogeneity; in the former the intensity is virtually unaltered
but the range changes. It is vital to account for this, far more so than for x-
ray therapy where combining beams can lead to tissue inhomogeneity effects
being no more than a few percent. For proton therapy the presence of a range-
shortening tissue inhomogeneity would reduce the target dose from the value in
the stretched Bragg peak to essentially zero (figure 4.3,which would be extremely
serious (Goitein 1982). Goitein (1982) identified three phenomena caused by
tissue inhomogeneities: range modification, edge scattering (if the inhomogeneity
does not fill the beam; see section 4.4.3) and dose modification from thin slivers. It
tums out that if slivers are small, multiple scattering wipes out their effect, which
is fortunate because otherwise impossibly accurate CT data would be required
for treatment planning. The range problem was recognized before the advent
of commercial CT and the group at LBL considered reconstructing from proton
projections to get the necessary CT images (Goitein 1972).
Since proton therapy is indicated when the highest conformation is needed,
precise patient positioning is required and patients are generally immobilized
comfortably in face masks (for head therapy) or shells (for body therapy). Goitein
et al (1982) report the amazing accuracy of 0.1 mm of the Lawrence Berkeley
immobilization devices. HCC-MGH assessed their mean error to be some 0.6

Pmton-therapyfacilities 179
to 0.9 mm. Historically, stereotactic devices have been used (see chapter 3) for
proton neurosurgery (Larsson et a! 1958).
Proton therapy is one of the techniques which, if more widely available, will
increase precision in radiation therapy, along with advances in three-dimensional
planning, on-line portal imaging, multileaf collimation and dynamic photon
therapy. Suit and Verhey (1988) have placed these new developments in context
with older developments (such as supervoltage radiation, portal radiographs,
simulators, isocentric gantries, secondary collimation) which over the last 50
years have transformed radiotherapy. They write, prophetically, 'It is to be hoped
that, by the year 2000, most of the treatment methods of 1987 will be judged as
obsolete'.
4.2. PROTON-THERAPYFACILITIES
As early as just after the second world war, Wilson (1946) wrote a historic paper
fortelling the use of protons in radiotherapy. It is instructive to reproduce a
calculation (in modem units) which he made to evaluate whether the proton beam
current would be large enough for medical purposes. In the last centimetre of
range a proton loses 30.1 MeV of energy in tissue (water) (put another way, the
range of a 30.1 MeV proton is 1 cm). A flux of F protons per cm2 deposits
F x (30.1 x 106)(1.6 x Joules (J) in 1 cm3. Assuming 1 cm of tissue
has a mass of 1 g, this energy deposition is F x (4.8 x lo-') J kg-'. So to create
a dose of 10 Gy (10 J kg-') requires 1010/4.8protons. Since each proton carries
a charge 1.6 x Coulombs, a 1 s exposure requires a proton beam current of
3.33 x lo-'' A. Wilson noted that the machines under construction could easily
achieve this and persevered with advocating proton radiotherapy.
Worldwide, the number of clinical proton therapy centres is not large (14 centres
in eight countries) (Chen and Goitein 1983, Slater et a1 1988, Bonnett 1991,
Sisterson 1990, 1991, 1992). The centres (see table 4.2) are at the Harvard
Cyclotron CentreMassachusetts General Hospital (HCC-MGH) (where by far
the largest number of patients in the world have been treated), at the Lawrence
Berkeley Laboratory-University of Califomia at San Fransisco (LBGUCSF), at
the Loma Linda Medical Centre, at the Gustaf Vemer Institute in Uppsala, and
at three centres in the (former) USSR, one in Switzerland, one in the UK, two in
France, one in Belgium and two centres in Japan.
The first proposal for the medical use of protons was made by Robert Wilson,
the founder of Fermilab, just after the second world war (Wilson 1946). He
proposed utilizing the accelerators which were being developed for high-energy
physics research. The first patients were treated in 1955 at the Lawrence Berkeley
Laboratory (Tobias er a1 1956, 1964), soon followed in Uppsala, Sweden in 1957
(Larsson et a1 1963, 1974). The HCC-MGH began treating patients in 1961,
although large-field irradiation did not begin until 1974 (Suit et a1 1975). Proton
therapy began in Dubna, (former) USSR in 1964, Moscow, (former) USSR in

sl8 %
E
%-
s
B E % E% %
% 2 z8 % >
3 S S%0 5 % ; %Z > 0
0
SE
s
2 h
3?! zO -22 F
2 2 z%
pI
Z
g 2 p
,
3 -5
~
m N8
m
*
5
E 2
0
0 3 -
2c 322 2 2
w m
.r +G
m sss v,
% P F % 3 Q 1 - e
o\m
gg
- I
03"
i
g,
2
5 zzs % s 2 2 2 2zz s2
2
--e
\ c
B
-r
r,
0 E
I
0
U 3
e
-e
e
0 c e o
g o e e $ 0
$
E -s -5 & & c -E l e a
U" 4-
0
5 g % g g" 3S
0 0 0 0
8 8
c) hh,ox e g g
g o . s t
U
.
6 Z cE ce ce E '*: &g
P c
- g O $ 2 2 2 8h
0
2 8-" 92 . eh 3 g
8
$ a
42
L
u g & $ 3 c 2 v l hv) v) v) VI
K9.i a
V V I -
-s
3 e!
VI .-
d
B <e! z
e!
2 bo
4
E d w z 3 h
-9
U
gE s
9 E L y e s
4.g l:-g -4 .g 3 s 5
;
a;.
.
c w
\I! P
B 4 g€ &TJ 3 w2
&&z"
3 2 2
0 8
2 %
.g >; &.a""- 2 4 . 2 E&
2
3
.B2.
-gz s
38 &5%
a z u n
3 { s g gg 33 .%
E
;
5 e
B S
$
24
E~
m' c"
0
'>
2
c"
S E 5
5 3 3
$8
L
:B m
2

2
$
Y
i5
3
E -
6
fi
G
73
V 2
.y
*
4.22- 3c - ."
e a $ Y
a
0
a 20 3 . s3 3
h
? mk
m
k
E
a
25.2 2 '=& .z
3 k0 c
Q gz 3 . ; 3 0 c g * ; ; ; . z a
v) 8
-1z
L '5 .& 8 .s 32
e
il
2O zm E0
*U $ b ~ 3 2 ~ ~ ~ 2 sp
.-~ ~ .
a 7 3 3 g 2 h k & c d v
.-8 4s %
, . E$ v x a3 ,% *.-
O $ . S $ e' &
= o e!u
a .=cP) U -g 3 . 2 2 a
m L c 'r .- .- ea < .g
'13 g 3 3 2 % 9 3 2 2 073 G % , g e A r g b) b)
&
P*
a2
-
c)
m
5
56 2 % 3 s
$,> t; -f t z -.sa 8 5a 0 0 5t: %
t; ,3s + .+z
' =g 2 , ~ + 3 g E
Fs 3 m Z a
4 3 U za ' s 8 -g az Y o Ze! 2 s m 2 - G 3 3-3
9 -
0
U" s
Proton-therapy facilities 181
1969 and at Gatchina, (former) USSR in 1975 (Ambrosimov et a1 1985). In Japan

proton therapy began at Chiba in 1979 and in Tsukuba in 1983 (Tsunemoto et a1
1985, Kitigawa 1988, Tsunemoto 1988). Swiss experience commenced in 1984
(Blattman 1986). A facility opened at the Clatterbridge Hospital, UK in 1989
(Bonnett 1991), at Loma Linda University in 1991 (Slater et a1 1990), at Nice in
1991 (Sisterson 1992), at Orsay in 1991 (Sisterson 1992), and at Louvain IaNeuve
in 1991 (Sisterson 1991). The worldwide clinical experience is summarized in
table 4.3. The main source of these data are the twice-yearly reports issued by
Sisterson at the HCC-MGH called the ‘Particles’ newsletter. Some starting dates
should not be interpreted too strictly as some of the above papers suggest slightly
different dates.
Proton radiotherapy began at Uppsala in 1957 using a 185 MeV synchrocy-
clotron. Protons of this energy have a range in tissue of 23 cm. Only a very
small number of patients were treated, generally those with advanced cancer or
where the tumour was very close to sensitive structures. The programme stopped
in 1968. The pencil beam was swept in a reticular pattern by two variable crossed
magnetic fields giving a transverse homogeneity of k 5 % over a 20 by 20 cm
field (Graffman et a1 1985). The Bragg peak was spread out by first a variable
water absorber, replaced by ridge filters which introduced different thicknesses
of absorbing material into the beam. Beams of cross section 0.1 to 20 cm in di-
ameter were achieved with dose rates (depending on field size) between 1 and
.
100 Gy min-’ The Uppsala team worked closely with all three proton acceler-
ator sites in the (former) USSR and shared experience. In 1985 plans were well
advanced for a new cyclotron facility in Uppsala with a variable energy acceler-
ator and an isocentric gantry and the first patients were treated in 1988 (see table
4.3). Clinical experience has been reported by Montelius et a1 (1991).
HCC-MGH has a 160 MeV synchrocyclotron with two horizontal beamlines,
one of 5 cm diameter with a doserate of 8-10 Gy min-’ and the other of 30 cm
diameter with a dose rate of 0.8-1.5 Gy min-’. The machine was constructed
between 1946-1949 (Blosser 1989) for physics research and started medical
therapy in 1961. The patient is arranged on a couch with precisely controllable
movements with a precision of &l mm. The patient is either sitting, standing
or decubitus. The treatment room is equipped with three orthogonally arranged
diagnostic x-ray sets for positioning. Figure 4.6 shows the geometry of the broad-
field proton beam. A double-scattering system with an annular occluding plate
can give a uniform dose across a 30 cm field width (figure 4.7) (see also section
4.3.2). The range modulator wheel (see section 4.3.1) is constructed of lucite,
different wheels being used to spread out the peak from 0.5 cm to 16.3 cm at the
90% isodose level (Sisterson and Johnson 1985, 1986 and 1987, Sisterson 1990).
The proton facility at the Institute for Theoretical and Experimental Physics
(ITEP) in Moscow is equipped with a 10 GeV proton synchrotron from which
protons can be extracted with energies between 70 and 200 MeV for medical
purposes. Only one in sixteen pulses are extracted by a beam-kicker, so medical
work can go on without interrupting high-energy physics activity. The accelerator

Table 4.3. Clinical experience to date with proton therapy facilities (table condensed
from Slater et a1 (1988), Bonnett (1991) and Sisterson (1990,1991, 1992)). This table is
updated every six months in ‘Particles’,newsletter of PTCOG (January and June). This
table is as of January 1992.
Location and Patients Year

Installation Country treated started
Lawrence Berkeley Laboratory Berkeley, 30 by 1957 1955

University of California CA, USA
Gustav Wemer Institute Uppsala, 73 by 1976 1957
(2nd installation) Sweden 23 by 1991 1988
Harvard Cyclotron Laboratory Cambridge, > 3000 by 1986 1961
MA, USA 4841 by 1990
5419by 1991
Joint Institute Dubna, 84 by 1980 1964
for Nuclear Research (former) USSR 13 by 1991 1987
(2nd installation)
Institute of Theoretical Moscow, 1250 by 1986 1969
and Experimental Physics (former) USSR 1993 by 1990
2200 by 1991
Konstantinov Institute Gatchina, Leningrad 381 by 1985 1975
of Nuclear Physics (former) USSR 719 by 1991
National Institute of Chiba, 29 by 1984 1979
Radiological Sciences Japan 74 by 1991
Particle Radiation Tsukuba, 158 by 1990 1983
Medical Centre Japan 242 by 1991
Swiss Institute for Villigen, 121 by 1986 1984
Nuclear Physics Switzerland 719 by 1990
1150by 1991
Clatterbridge Hospital Clatterbridge, 83 by 1990 1989
UK 189 by 1991
Loma Linda University, Loma Linda, 76 by 1991 1991
Califomia USA
Universit6 Catholique Louvain la Neuve, 8 by 1991 1991
de Louvain Belgium
Centre Antoine Lacassagne Nice, 45 by 1992 1991
France
Centre de ProtonthCrapy Orsay, 13 by 1992 1991
France
Total proton beams worldwide: 10 358 by end 1991

Proton-therapy facilities 183
-- FIRST SCATTERER
I 3 c m APERTURE
GEOMETRY OF
SECOND SCATTERER
ANTI-SCP;TTER
CONCRETE
Figure 4.6. Schematic of the spread-out proton beam, designed by the

Harvard Cyclotron Laboratory, showing the location of the scattering foils in
the passive scattering system, lucite range-modulator wheel and the patient’s
beam-defining aperture. The individual patient compensator (bolus) would
be placed after this aperture. The beam is monitored by an ion chamber. The
scale of this diagram is distorted for convenient display (see key bottom left).
(From Sisterson and Johnson (1986).)
feeds three treatment rooms equiped with stereotactic localization apparatus for
neurological procedures. Lasers and x-ray based positioning equipment is also in
use. The beam energies are degraded and large fields produced by scattering in
thin foils (see section 4.3.2). Dose planning would appear to be by hand with a
‘dose field catalogue’ created by dosimetric instruments, claimed to be accurate
to 7-9% (Chuvilo et a1 1984, Khoroshkov et a1 1969).
The proton facility at the National Institute of Radiological Sciences, Chiba,
Japan has a cyclotron-generated 70 MeV beam only suitable for superficial
radiotherapy, penetrating 3.8 cm in water. The Bragg peak is only 1.3 mm wide
and is stretched by a rotating plastic degrader to 2.5 cm. The proton beam is

The physics of proton radiotherapy
EXA"CD,E Or -f?EfiTlrlEriT EL0
0
- 8 6---12 0 - 8 - 6 - 4 - 2 C 2 4 6 8 0 2 _I D 8 -
TR*'bS+E9SC S C f h eT CT CEPTLi 1E.l +?l;-EF
Figure 4.7. A transverse scan in a water phantom for the 160 MeV proton
beam at the Harvard Cyclotron Laboratory showing the uniformity of the
beam and the sharp lateral fall-off of the dose (From Sisterson and Johnson
(1986) )
defocused electromagnetically and collimated mechanically into a 1 cm2 field.

This is swept horizontally and vertically with two perpendicular swing-magnets
to create a uniform field up to 20 by 20 cm2.
Proton therapy is practised in the UK at the Clatterbridge Hospital (Bonnett
1991) where, being a 'low'-energy facility it is used exclusively for treating
occular melanoma. There are plans to introduce an 'add-on' linear accelerator
to boost the energy to 180 MeV, due to be completed in 1994 (Sisterson 1992).
A proton therapy facility has just been completed at Loma Linda University
(Antoine 1987, Awschalom et a1 1987, Miller et a1 1987, Slater et a1 1988, 1990,
1992) having a 70-250 MeV synchrotron directed to one stationary beam-line and
three rotation gantries. Each gantry can rotate the beam through a full revolution
about a patient on a couch as in x-radiation therapy. The synchrotron is the world's
smallest, built by Fermi National Accelerator Laboratory; it is a ring of magnets
some 20 feet in diameter through which protons circulate in a vacuum tube. The
energy of the protons increases as the magnetic field is increased reaching 250
MeV in s. The beam is steered by giant magnets to gantries weighing 95 tons
and standing three stories high. These gantries are 35 feet in diameter supporting
the bending magnets and focusing magnets. All the patient sees is a small tunnel
through the facility. The details of the accelerator construction and beam transport
systems have been provided by Slater et a1 (1988). The horizontal beam-line
can also be operated at 155 MeV and 200 MeV and there are plans to produce
a continuously variable energy beam in steps of 5 MeV (Sisterson 1992). This
is a major investment in technology; as well as the proton facility, all ancillary
imaging equipment (CT, PET and SPECT) and no less than eight treatment planning
workstations are installed (Potts er a1 1988). The Loma Linda facility achieves
2.5 x 10" protons per spill, a repetition rate of 30 spills min-' corresponding to
50 cGy min-' for a 20 cm diameter area and 9 cm range modulation (Coutrakon
er a1 1991a). The first occular treatment was on October 23rd 1990 (Slater et a1

Proton-therapyfacilities 185
1992). Treating brain tumours began in March 1991 and treatments began with
the rotating gantry beam-line on June 26th 1991. 76 patients had been treated by
the end of December 1991 (Sisterson 1992).
The history of the Loma Linda facility has been published by Slater et al
(1992) including an account of how the Proton Therapy Cooperative Group
(PTCOG) formed. Loma Linda is unique in that the building and accelerator were
planned together for the efficient use of both. The structure of the building even
accommodated for the relatively high seismic activity in Southern California.
The latest proton therapy installations include Louvain la Neuve in Belgium
(Sisterson 1991). It has a proton energy of 90 MeV with a maximum water
penetration of 5.5 cm. The first treatment was in January 1991 and eight patients
had been treated by September 1991 at a dose rate of 1-2 Gy min-’. The proton
facility in Nice has a 65 MeV cyclotron-generated beam. Treatments started on
June 17th 1991. The second French facility at Orsay has a synchrocyclotron and
started treating patients on September 17th 1991 (Sisterson 1992).
Most proton accelerators used for radiotherapy are cyclotrons or synchrotrons
(Beeckman et a1 1991, Gottschalk 1987). Variations for cyclotrons include
the synchrocyclotron and the isochronous cyclotron with magnets at room
temperature or superconducting (Blosser 1989). The synchrocyclotron is a
particularly reliable machine; Sisterson et a1 (1991) have quantified the relatively
small ‘downtime’ at the HCC machine. A room-temperature cyclotron is too
massive to rotate and to provide a rotating beam, some beam-transport mechanics
are needed, as in the design by Beeckmann et a1 (1991). Recognizing that the
ability to rotate the proton beam about a stationary patient is highly desirable,
Blosser (1989) has designed a superconducting isochronous cyclotron which
could be rotated about the patient as shown in figure 4.8. The magnets in
superconducting cyclotrons are some 17 times less heavy than their room-
temperature counterparts for the same bending power, allowing the cyclotron to
itself rotate about the patient. This removes the need for the elaborate beam
transport needed for the rotating gantry for room-temperature cyclotrons or for
the proton synchrotron as at Loma Linda.
Recently Lennox (1991) has proposed that a single proton linear accelerator
could be used with the appropriate beam-switching techniques to achieve the
following four purposes at the same clinical site: (i) proton therapy, (ii) fast-
neutron therapy (iii) boron-capture neutron therapy (using epithermal-1 keV-
neutrons from 70 MeV protons striking a lead target surrounded by an iron
moderator) and (iv) isotope production for PET imaging.
Features of the other existing proton facilities are discussed elsewhere in
this chapter (section 4.3). There are many proposals for future proton therapy
installations. These have been summarized by Bonnett (1991) and Sisterson
(1992) and include the British proposal by AEA Technology (Oxford UK) for
a linac-based facility delivering protons of energies 100, 150, 200 and 250 MeV.
Sisterson (1992) has recently reported pre-clinical measurements at the Indiana
University Cyclotron Facility proton beam-line (185-200 MeV).

SUPPORT J LROTATING
SHIELDING DISK
ROLLERS
Figure 4.8. Showing a likely layout for a one-treatment-room proton

therapy facility. The cyclotron and beam-transport system rotate as a unit
so that intricate out-ofplane bending of the beam is not required. Thefigure
shows an optional wobbler magnet, although moving the patient relative to
the beam which has been focused into a perpendicular vertical line is more
attractive. The isochronous cyclotron has a superconducting magnet which
reduces weight compared with its room temperature counterpart and allows
the rotation. (From Blosser (1989).)
4.3. RANGE MODULATION AND PRODUCTION OF LARGE-AREA

BEAMS
The spreading out of the Bragg peak or range modulation can be achieved in a
number of ways (Bonnett 1991, Graffman et a1 1985, Goldin and Monastyrsky
1978, Koehler er a1 1975). The Bragg peak is spread out by introducing extra
absorbing material before the beam enters the patient. If different thicknesses of
such absorber are present for different fractions of the irradiation time, the narrow
monoenergetic peak can be spread into a useful plateau. This allows the high-dose
deposition region to straddle the target volume containing the tumour. Ways to do
this are described in section 4.3.1.
As well as spreading the peak, the final range itself must be shaped to the distal
surface of the target volume. Bonnett (1991) sums up ways to do this:
1. Fixed range modulation (see section 4.4.2): a compensator is constructed to
shape the distal surface of the target volume to the maximum proton range.
2. Variable range modulation and spot scanning: these techniques (Chu et a1
1989) to spare normal tissue proximal to the target volume are under development.
The production of a large-area flat beam can be achieved by (Chu et a1 1989):

Range modulation and production of large-area beams 187
1. Scattering from foils (see section 4.3.2): this was the first way to be used
clinically (Koehler et a1 1977) and is still the main method in use (see figure 4.6).
The main disadvantage is the energy loss in scattering and the reduction of beam
output intensity. The penumbra is also broadened.
2. Pixel or spot scanning (see section 4.3.3): the target area is divided into
pixels, scanned separately. The scanning method does not degrade beam energy
nor output. The sharp penumbra is retained and large field sizes are attainable.
4.3.1. Spreading out the Bragg peak

The Bragg peak can be spread out to a useful plateau by the use of a rotating
stepped absorber. The idea was first suggested by Wilson (1946) and Koehler et
a1 (1975) provide details of the implementation at the HCC-MGH proton facility.
The idea is elegantly simple. It relies on time modulation, presenting different
thicknesses of absorber to the beam for different fractions of the irradiation time
(figure 4.9). A range modulator wheel (figure 4.10) is constructed from a pyramid
of blades each in the shape of a fan, of different thicknesses and different fan
angles. The thinnest part of the stack has the widest fan angle. The modulator
rotates at 90 RPM in the beam-line (figure 4.11).
The thickest part of the modulator pulls back the Bragg peak furthest. A
computer program determines the amplitudes of elemental Bragg curves with
L depth
Figure 4.9. Shows the principle of range modulation to produce the

spread-out Bragg peak (SOBP). Curve ( a ) shows the depth dose of an
unmodulated proton beam. Curves (b)and ( c ) show the effect of introducing
absorber upstream of the patient. As the thickness of the absorber increases,
the Bragg peak becomes more proximal to the surface. Curve ( d ) is an SOBP
created by the range-modulator wheel (shown end on schematically). The
small triangles indicate the fraction of the time for which each thickness
of absorber should be in the beam. The triangles are at the depths of the
elemental proton ranges. This in turn determines the differential fan angles
for the rotating wheel.

Figure 4.10. The 14.5 cm-range, rotating range-modulator wheel of the

HCC-MGH proton facility. The wheel is constructed from a pyramid of
plexiglass fans of different opening angle. The wheel is 82 cm in diameter
and the circular beam-definingaperture visible through the wheel is 27 cm in
diameter. (From Koehler et a1 (1975).)
Figure 4.11. Layout of the proton-beam treatment area at the HCC-MGH

proton ,facility showing (left) the range-modulator wheel mounted on to
the collimator. An arrangement is shown on the right for measuring the
depth-dose curve. (From Koehler et a1 (1975).)
different ranges which, when added, would give the required plateau. These
amplitudes are then converted to a set of differential fan angles, normalized so
the integral fan angle (the sum of the differentials) is 180". The differential fan
angle is that part of 180" for which a particular thickness of modulator is in the
beam. From this set of differential fan angles the integral opening angle or fan
angle of that blade corresponding to the ith Bragg sub-peak is then the sum of

+
the integral fan angle for the ( i 1)th Bragg sub-peak and the differential fan
angle for the ith Bragg sub-peak (the sub-peaks are labelled so that '0' labels the
unmodulated peak and the label increases in the direction of decreasing range or
increasing thickness of modulator).
The modulator is normalized to 180" rather than 360" so that each blade appears
twice, diagonally opposed. This ensures dynamic equilibrium as the modulator
rotates. As can be seen, part of the time the beam passes unmodulated. The HCC-
MGH wheels (of which there were three for different plateaux) were constructed
from plexiglass, a low-Z, little-multiple-scattering material. The wheels were 82
cm in diameter.
Measurements on the Loma Linda proton facility showed the range of the
unmodulated proton beam was 32.7 cm which, taking account of material
upstream of the measuring water tank, corresponded to a beam energy of 235
MeV. When 5 mm of lead were upstream, the range shifted to 29.5 cm. In both
cases the peak-to-entrance dose was approximately 3: 1 and the distal edge fall off
from 90% to 10% was 7 mm. A propeller, rotating at 310 RPM, to spread out the
Bragg peak was designed by the Harvard Cyclotron Medical Facility for Loma
Linda, comprising nine sections of lucite, each 8 mm thick reducing the range by
1 cm of water. This created an SOBP of width 9 cm with uniformity f5%. The
distal dose fall-off was then 8 mm with a distal edge at a depth of 27 cm in water,
relecting the range reduction introduced by the scatterers (see section 4.3.2). The
peak-to-entrance dose ratio for the SOBP fell to 1.3:l (Coutrakon et a1 1991a).
The Indiana University proton beam-line uses a similar mechanism to produce an
SOBP with a 1.3% uniformity (Sisterson 1992).
The 'propeller' method of creating an SOBP is applicable to a beam which has
been spread out by a double-scattering-plus-annulus arrangement. It could be used
with a small-area beam but generally is not. When a small-area proton beam is
used to create a large-area field by spot scanning, another method of spreading out
the Bragg peak is needed. In some installations this is achieved by rapidly shooting
the appropriate thickness of absorber into the beam under pneumatic control.
Another very elegant method which has been developed by the company Ion
Beam Applications (IBA: Belgium) is to make use of a rotating hollow cylinder of
diameter 20 cm with variable wall thickness. The beam is directed at right angles
to the long axis of the cylinder and is much narrower than its diameter. The beam
intersects the axis of rotation of the cylinder. The walls of the cylinder increase
from some minimum thickness to a maximum thickness diametrically opposite,
retuming immediately to the minimum thickness and thereafter increasing again
to the maximum, 180" further round. This cylinder rotates at some 1500-3000
RPM, presenting a variable thickness of absorber which is precisely known as a
function of the angular orientation of the cylinder. This orientation is monitored
by an optical encoder which can give signals to the beam control so that the beam
is only switched on for specific chosen parts of each rotation. If the beam is on
continuously then the dose in the patient will exhibit the maximum width of SOBP.
If the beam is switched off for a selected portion of the rotation between when the

cylinder presents some thickness of wall greater than the minimum up to when
the maximum thickness is presented (corresponding to removing the contribution
from selected lower energy protons), the width of the SOBP can be reduced. In this
way the width of the SOBP can be controlled. The distal extremum of the SOBP
is separately vaned by a stationary absorber upstream towards the isochronous
cyclotron (Jongen 1992a,b, Jongen et a1 1992).
4.3.2. Theory offlattening proton dose distributions for large-jield radiotherapy

When a thin foil is placed in a narrow beam of protons the resulting projected
distribution on the treatment, or on some measurement, plane a distance ZI from
the foil is given by the radially symmetric function of radius rl :
fPI) = (VxR:) exp ( - r M ) (4.2)
where R I is the root-mean-squareradius of multiple scattering, related to the root-

mean-square scattering angle (Of)by
R~ = zl(e;p”. (4.3)
The multiplicative constant in equation (4.2) normalizes the distribution such that
1f (r1)h1= 1. (4.4)
4
This is easy to prove if is written as x2+y2, the integral is split into two integrals
over x and y , variables are changed to t = x / R 1 and t = y/R1 and we recall that
Jdm exp dt = f i . (4.5)
The proton distribution given by equation (4.2) is far too centrally peaked to
be useful for wide-field irradiation and is too non-uniform. Figure 4.12 shows the
Gaussian distribution. Koehler et a1 (1977) pioneered techniques for flattening the
beam as follows. A second scatterer was placed downstream from the first along
with a beam-stopper on the central axis. The purpose of this beam-stopper is to
expose the second scattering foil to just part of the first-scattered beam and this
creates the flat distribution, as will be seen in a moment. Photons which interact
on the second scatterer at positions which project to rl on the measurement plane
give rise to a distribution f (r2),where r2 is the vector from the position rl (see
figure 4.13). The final distribution as a function of radius r at the measurement
plane is then
f ( r ) = f (r1)f (r2) . (4.6)
Vectorial1y
r=rl +r2 (4.7)

A
FIRST SCATTERER SECOND SCATTERER MEASURING P I A N E
APERTURE STOPPER
Figure 4.12. The geometry ofdouble scattering with a beam-stopper infront

of the second scatterer. ( A ) shows the defining aperture and the two scattering
foils. The beam-stopper projects to a radius A1 at the measurement plane.
( B ) shows the beam profiles ( I ) after the first scattering foil, (2) after the
beam-stopper, (3)at the measurement plane (where the distribution has been
jattened; see figure 4.7).(From Koehler et a1 (1977).)
i.e.
r i = r2 + r; - 2rrl cos8 (4.8)
and so combining equations (4.2), (4.6) and (4.7)
This is the probability of scattering into a radius r at the measurement plane

Z1 via a projected radius rl from the first foil, where rl are all the positions on
the circumference of a circle with this radius. The integral over 8 performs this
task and the 2 on the numerator is because the integral should be over 2 x but is
symmetric so can be reduced to half the range. The parameter R2 is given by
(analogous to equation (4.3))
R2 = 22(8i}1'2 (4.10)
where (8i)'/2is the RMS scattering angle for the second foil. The two scattering
angles depend on the material and thickness of the foil. Once these are known, RI
and R2 follow from the geometry.
Now make the substitutions
x = rl/R2 (4.11)
(4.12)
and
(4.13)

A B
Figure 4.13. The projection of scattered rays from the two scatterers on
to the measurement plane at Z1from the first scatterer. The left diagram (A)
shows a solid beam-stopper. Valuesof RI inside A1 are forbidden butprotons
scatter from values rl A1 back inside A I to fill up the space. The right
diagram ( B ) shows an annular beam-stopper. Here the disallowed values of
rl are between A I and A2. Small values of r2 from rl fill in the shadowed
region (seefigure 4.12). (From Koehler et a1 (1977).)
to obtain
f (r,x ) = -exp
TT R: R;
(- (kix2+ 5)) (I/*) 0
exp ( k 2 x COS 0 ) de.
(4.14)
But by definition of the Bessel function of an imaginary argument
Io ( 2 )= ( l / n ) lr
exp ( 2 cos e ) de (4.15)
To obtain the distribution over r alone we need to integrate over the whole area
in which rl (i.e. in which x ) can lie. So we need to integrate with respect to x dn
to perform the integration over the slim annular area bounded by circles of radius
x andx +dx. Thus
This is the complete expression for the two-dimensional function f (r). It is given
in terms of just two parameters R I and R2, which, as we have seen, depend on the
scattering foil.
The key to understanding the importance of Koehler et a1 's (1977) development
is recognizing that the remaining free choices are the limits of the integral. These
limits are in terms of x or (from equation (4.1 1)) r1, the projected position of the
first scatter at the measurement plane. If a radially symmetric beam-stopper is
placed centrally in the beam just in front of the second scattering foil (figure 4.12)

Radius ( i n u n i t s of A , )
Figure 4.14. A calculated scattered distribution of protons with a central

beam-stopper of projected radius A1 when ‘good parameters R I = 1.7A1
and R2 = 1.3Al were chosen (curve ( a ) ) . When ’poor’ parameters were
chosen, R I = 1.7A1 and Rz = 1.OA1, the distribution is notflattened (curve
( b ) ) . Curve (c) shows the total dose within a given radius. The beam-stop is
shown hatched bottom left. (From Koehler et a1 (1977).)
this disallows a range of values for rl (and so for x ) . Two types of beam-stopper
were proposed by Koehler et a1 ( 1 977):
1. A single central stopper of radius A 1 (projected at the measurement plane)
blocking the central portion of the beam (i.e. rl > A I only). The stopper was a
brass block, tapered to the divergence of the beam and of thickness 3.7 cm which
completely stops protons of 160 MeV (figures 4.12 and 4.13).
2. An annular stopper, open to the central portion but stopping the beam
between rl values of A 1 (projected outer radius) and A2 (projected inner radius).
This was used for very large fields (figure 4.13).
Figure 4.14 shows a calculated scattered distribution of protons with a central
beam-stopper of projected radius A1 when ‘good’ parameters RI = 1.7A1 and
R2 = 1.3A1 were chosen. Notice the beam is above 95% out to about 1.5 A I
(curve (a)). When ‘poor’ parameters were chosen, R I = 1.7A1 and RZ = 1.OA1,
the distribution is not flattened (curve (b)). Curve (c) shows the integrated dose
within a given radius.
Figure 4.15 shows a calculated scattered distribution of protons with an annular
beam stopper of projected radii A I and A 2 when ‘good’ parameters RI = 1.7A I
and R2 = 0.8A1 were chosen and A 2 = 0.38Al. When ‘poor’ parameters were
chosen, R I = 1.7A1 and R2 = 0.7A1 and A 2 = O S A I , the distribution is not
flattened (curve (b)).Curve (c) shows the integrated dose within a given radius.
The first blocking technique with a single stopper was used at HCC-MGH from
May 1973 for intracranial targets. The second technique with the annular blocker
was used from December 1973 for large-field irradiations. Koehler et a1 (1977)

R a d i u s (in u n i l s o f A , 1
Figure 4.15. Shows a calculated scattered distribution of protons with an

annular beam-stopper ofprojected radii A 1 and A2 when ‘good’parameters
R1 = 1 . 7 A 1 and RZ = 0.8Al were chosen and A2 = 0.38A1 (curve ( a ) ) .
When ‘poor’ parameters were chosen, R1 = 1.7A1 and R2 = 0.7A1 and
A2 = O S A I , the distribution is notjattened (curve (b)).Curve (c)shows the
total dose within a given radius. The beam-stop is shown hatched bottom left.
(From Koehler et a1 (1977).)
christened these systems ‘proton nozzles’ and they have proved very effective.
The double-scattering-plus-annulusmethod was also used at the newest proton
facility in Belgium (Sisterson 1991) giving field sizes up to 10 by 10 cm2, at
Tsukuba, Japan (Tsunemoto et a1 1985)and at the Indiana University proton beam-
line (Sisterson 1992).
A double-scattering technique similar to this has been installed at the Loma
Linda proton facility. The first foil is 6.75 mm of lead and the second foil is 0.5 mm
of lead on a in lucite plate. The second scatterer was immediately followed by
two concentric brass occluding rings of diameters 0.77 to 1.45 cm and 2.4 to 3.05
cm and total thickness 7cm. Measurements with both a small silicon diode and
with a 400-element ion chamber showed the proton flux was uniform to within 2%
over a 20 cm diameter area, but that this uniformity depended on good alignment
of the occluding rings (Coutrakon et a1 1991a).
At the National Institute of Radiological Sciences (NIRS), Chiba, Japan, a
quite different method of achieving large uniform fields is used. The elemental
proton beam is some 1 cm2 and this is swept horizontally and vertically with two
perpendicular swing magnets to create a uniform field up to 20 by 20 cm’ (Kanai
et a1 1991). We now tum attention to the altemative method.
4.3.3. spot scanning

The method of spot scanning has been used at the NIRS 70 MeV proton facility at
Chiba, Japan (Kanai et a1 1980,199I, Tsunemoto et a1 1985, Kawachi et a1 1983).

The beam from the cyclotron is shaped to a 10 mm by 10 mm square field by a

series of apertures and magnets. Range modulation is achieved by a rotating lucite
disk of variable thickness. The elementary beam is moved in space by horizontal
and vertical scanning magnets capable of creating a field of size 18 cm by 18
cm. When the elementary fields were all of the same duration, the uniformity
of the large field so created was 2.5%. The position of the elementary field was
switchable by 1 cm in 1 ms. Hence to achieve 2% accuracy the exposure time at
any particular location should not fall below 50 ms. In practice the beam delivered
some 1 Gy s-l over a lcm by 1 cm field, so it took about 3 minutes to scan a 10
cm by 10 cm field delivering 2 Gy.
The advantage of spot scanning over the use of scattering foils is that highly non-
uniform irradiation pattems are achievable. Figure 4.16 shows such a complex
irradiation field and how well this can be achieved. The scanning pattem is
prestored into a computer along with the time which the beam should spend at
each location. A multileaf collimator comprising 40 1 cm2 brass rods provides
back-up collimation. Tsunemoto et a1 (1985) report on a fully three-dimensional
spot-scanning technique. Each CT slice is treated separately and within the slice,
the beam is scanned incrementally. At each position within the slice a different
energy degrader was placed in the beam to tailor the Bragg peak to the local distal
edge of the target.
It is important to realize that this method of spatially varying the proton
beam intensity is quite different from how an x-ray field can be modulated with
compensators. For protons, compensators shift the range of the protons, not their
intensity. To modulate the proton intensity, the duration time has to be modulated.
The raster-scanning method of creating large-area fields can also used for the
proton facility at Loma Linda (Awschalom et a1 1987,Miller et a1 1987, Coutrakon
et a1 1991b). The elementary beam is 2 cm in diameter and can be swept to fields
as large as 40 by 40 cm at a maximum sweep speed of 2 cm ms-’.
Phillips et a1 (1992) report the spot-scanning method at the Paul Scherrer
Institute (PSI), Villigen, Switzerland. The small-area beam can be steered in the
( x y ) plane at right angles to its direction by sweeper magnets. To move the Bragg
peak in the patient along the direction (z) of propagation of the beam, range-shifter
plates of different thicknesses can be inserted. The order of scanning is x , z, y.
This method of covering the three-dimensionalvolume of a target would allow the
intensity of the beam to be varied as a function of the three-dimensional location
of the Bragg peak by methods such as that described by Webb (1989) for photons
(see chapter 2).
Respiratory motion complicates the technique and was extensively evaluated
by Phillips et a1 (1992). Respiration was simulated by perturbing the base
beamspot position by a sinusoidal motion whose amplitude, frequency and phase
could be varied. For single fractions, the dose inhomogeneity increased as
the amplitude of respiration increased as expected. By keeping the amplitude
and frequency constant and varying the phase, the effects of fractionation were
studied and shown to ‘smooth over’ otherwise irregular distributions of dose. The

60 -
E- 40-
5 20-
0
z 0-io) bt
:2 0 -
+
11
2 40-
60 -
L A I 1 $ , , , I I
60 40 20 0 20 40 60
Distance off axis lmn)
lo)
60 40 20 0 20 40 60
Distanceoff axis Imm)
I b)
Figure 4.16. (a) A two-dimensional modulation of proton intensity across a
large field created by a spot-scanning method. The dose to area 2 is twice the
dose to area 3. The dose to area 1 is zero. ( b )A profile across the line a-b of
the measured distribution of dose. (From Kanai et a1 (1980).)
percentage standard deviation decreased roughly as the square root of the number
of fractions. Respiration frequency (between 6 and 60 breaths per minute) had
little effect on dose homogeneity. There was a strong correlation between the
dose inhomogeneity and the relationship between the movement direction and
the direction of scanning. The fastest scan direction ( x ) should be chosen to
correspond to the axis of the largest respiration motion. The slowest scan direction
(y) should be along the direction of smallest motion.
4.3.4. Monitoring the beam intensity

Any beam-intensity monitor must satisfy the following design constraints:
1. it must present minimal mass to the beam in order not to degrade the energy
(and hence the proton range), nor greatly degrade the beam size (which would
affect the transport to the treatment room),
2. it must be radiation hard,
3. it should ideally measure radiation intensity to 1% accuracy in about 1 ms if
the beam is swept magnetically.

Beam L.* IO
Figure 4.17. The variable energy proton synchrotron for the Loma Linda
Cancer therapy facility. The air-core quadrupole in the accelerator regulates
the beam extraction using the beam intensity monitoring signal from the
treatment room. The intensity monitor is described in section 4.3.4. (From
Coutrakon et a1 (1991b).)
A beam-intensity monitor satisfying these requirements has been developed for

the Loma Linda facility (figure 4.17) (Coutrakon et a1 1991b). It comprises a 5
cm thick gas cell coupled to a photomultiplier tube. The gas cell was filled with
xenon at 1.2 atmospheres pressure with 12.7 g m thick titanium windows. This
presents 0.14 g cm-* total mass to the beam, degrading the energy by 70 keV
corresponding to a loss of 1.4 mm range in water. The beam increases in size by
6mm at the patient site after transport through 3 m, an extra angular spread of 1.88
milliradians.
Each passing proton produces about 2000 photons (70000 eV of energy are
deposited per proton; about 20% of the energy deposited in a noble gas is
dissipated radiatively, in this case as 7 eV photons). Of these, about 0.1% (i.e.
two photons per proton) are detected by the photomultiplier tube whose spectral
response overlaps the xenon scintillation spectrum by some 50%. The proton
beam can operate at upwards of 1O’O protons per 1 s ‘spill’. Thus in the 1 ms
that the beam is at some particular geometric location in its sweep, some lo7
photoelectrons are produced, which is adequate to measure the intensity to better
than 1% accuracy.

4.4. PROTON TREATMENT PLANNING
Goitein et a1 (1982) have reviewed all aspects of planning proton therapy. Of

particular note are:
1. the need to consider the full 3D nature of the problem,
2. the need to account for tissue inhomogeneities (more so than with photon
planning),
3. design of compensators,
4. special imaging procedures for planning proton therapy,
5 . exquisite patient immobilization,
6. the need to assess and plan for unknowns,
7. the need to compare plans with ‘rival’ photon plans.
4.4.1. Tomographic imaging for proton treatment planning

As for planning x-ray therapy, CT tomographic data are an essential pre-requisite
to proton treatment planning (Chen 1983). As shown above, the physics
determining how proton beams deposit their energy in tissue is different from
x-rays and the use made of CT images is therefore quite different. The major
concem is that tissue inhomogeneities adjust the proton range (not intensity) and
could if improperly handled lead to serious underdosage. The inability to accurate
map inhomogeneities prior to commercially available CT was a major hindrance
to its acceptability and development. Goitein (1977) has studied the physical
specification of the CT scanner. Consider in turn, density resolution and spatial
resolution.
At the end of its range the proton dose falls from 80% to 20% in about 4 mm
suggesting that an accuracy of f 2 mm is necessary in specifying the proton range.
If this range is say 20 cm, this implies a 1% error would be acceptable. If the pixel
size is say 2 mm (i.e. there are 100 pixels along the range) the random error per
pixel in CT number which leads to this overall error is 10%. This is considerably
larger than the 1% density resolution which CT scanners can achieve. However,
systematic errors must be kept to 1% and this is just at the limit of present CT
technology.
Now consider spatial resolution. The effect of a thin sliver inhomogeneity
whose thickness is less than the proton range critically depends on its width (very
thin slivers hardly cast any shadow because multiple scattering fills in the dose
distally-see figure 4.18). However, if the spatial resolution of the CT images
were poor (say 2 or 3 mm instead of the usual 2 1 mm) such slivers would appear
to be larger, and importantly, wider than they really are. If such low-resolution
images were used to plan proton treatment, the dose-perturbing effect of small
structures would be misjudged. The wider the inhomogeneity appears, the more
the range appears foreshortened (see figure 4.18 again). Goitein (1977) thus argues
that a spatial resolution of 1-2 mm is needed, fortuitously matched by the actual
performance of state-of-the-art CT scanners. Because the range-modifying effects

Proton treatment planning 199
I I I I
20
0 I I
0 5 10 15
DEPTH (cm)
Figure 4.18. The calculated central-axis dose distribution for an initially

parallel energy-modulated proton beam incident on a sliver of tejon of length
2.5 cm and different widths. The graphs show how, for thin slivers, multiple
scatter can fill in the shadow region, whereas this effect decreases as the width
( U J ) of the sliver increases. This has important implications for the required
resolution of CT images used for planning proton therapy. (From Goitein
(1977).)(Reprinted with permission from Pergamon Press Ltd, Oxford, UK.)
of inhomogeneities are not confined to two-dimensions, it is however important

that the axial spatial resolution in CT is also 1-2 mm. This is generally not the
case for diagnostic scanning so special arrangements must be made if CT data are
to be used for planning proton therapy.
High-Z material in the patient can present problems when using CT data for
planning proton therapy. The stopping power of protons is determined by the
electron density of tissues. For most soft tissues, it is the electron density which
maps to the CT number because the dominant mechanism is Compton interaction.
Hence for such tissues CT data can be used to predict the range directly. However,
the CT number is determined by the total linear attenuation coefficient of x-rays
( p in cm-'> at the effective energy of about 80 keV. The fraction of the total
attenuation due to Compton interaction falls off with increasing atomic number Z
(see figure 4.19). Also the physical density p in g cm-3 increases with increasing
Z . Hence the fraction of the total attenuation coefficient ~ / inpcm2 g-' due to the
Compton effect falls with increasing Z. Goitein (1977) plotted the dimensionless
product of the proton range in g cm-* and this fraction, which is the electron
component of linear x-ray attenuation (figure 4.19), and showed how it is therefore
not a horizontally straight line, as one would like, but varies with Z. However,
he argues that although this appears to be a problem, in practice the volume of
high-Z material in the patient is not large and can possibly be ignored to a first
approximation. Alternatively,dual-energy CT scanning can be used to address the
problem.

---.
/c
loo - c - - - - - c ~ ~ . ,
5 - c c
---_ _ _ - -
2% 7 ,,e
1_1;
\\
-
,
4 . ,’
50 - ?
3 , I 1
ATOMIC NUMBER
ATOMIC NUMBER
iai
(b)
Figure 4.19. ( a ) The proportion of x-ray attenuation coefficient at 80 keV

due to the Compton effect. ( b )The product of proton range and Compton not
total x-ray linear attenuation coefficient as a function of the atomic number
2 . Note that this graph is not horizontal because both the component of
attenuation due to Compton scatter decreases with increasing 2 (see ( a )
above) and also because of the 2-dependence of physical density. (From
Goitein ( I 977).) (Reprinted with permission j?om Pergamon Press Ltd,
Odord, UK.)
In conclusion, commercial CT scanners are entirely appropriate for proton

therapy planning, provided the slice width is reduced to 1-2 mm.
4.4.2. Theory of range modulation by compensator or bolus

Compensating for tissue inhomogeneities is a vital component of proton
treatment planning. The penetration of the beam must be modified to take them
into account. Where the integrated stopping power is large the beam must be made
correspondingly penetrating and vice versa. Computed tomography provides
the necessary data (section 4.4.1). Range modulation may be accomplished by
varying the maximum proton energy while scanning a narrow beam across the
field. Altematively, and more commonly, bolus applied extemally to the patient
achieves the range modulation. In principle the method is no more complicated
than applying, at each location in a field, the appropriate thickness of bolus to
force the distal dose edge (say the 90% dose contour) to follow the distal edge of
the target volume. The bolus is thick where the beam should be less penetrating
and vice versa. One consequence of this bolus is that the proximal 90% isodose
contour will now become a ragged surface in a heterogeneous medium. Figure
4.20 shows in two-dimensions how a compensator can account for a number of
factors to match the range to a prescription.
Figure 4.21 shows how to compute the compensator thickness. An element of
the beam passes a pre-absorber PA, a defining aperture A, through the compensator
C into the treatment volume V. For element ax,yof the compensator the thickness
is cXy.If the CT density of the patient is pXyzthen the required range in the patient
to reach a depth z,, is

p;
.................... ..,
;,'O
jp 1 ............ :
,'
ui
Figure 4.20. Shows how a range-compensator can adjust the proton
penetration so the range matches a prescription. The beam enters at the
left ana' passes through a k e d path-length absorber. The 'patient' is to the
right with the compensator in between. Beneath region A there is an organ at
risk (shown circular) so the compensator has a circular increase in depth
to pull back the range. In region B the patient external contour starts to
fall away so the compensator must increase in thickness. Beneath region C
there is a circular bone inhomogeneity which causes range shortening. The
compensator is correspondingly thinned. In practice all these effects may
occur together and the overall shape of the compensator would be as shown.
(From Wagner (1982).)
PA A C
V
Figure 4.21. Shows how to compute the compensator thickness. An element
of the beam passes a pre-absorber PA, a defining aperture A , through
the compensator C , into the treatment volume I! For element a,,, of the
compensator the thickness is cxy. The CT density of the patient is pxy. . ZQ
is the entrance depth of the patient ana' zmax is the required proton range (or
where the distal 90% contour should be). (From Wagner (1982 j.)
(4.18)
The absorber thickness should then be
(4.19)

where P is the maximum proton range in water (e.g. 16 cm at 160 MeV) and the
factor 1.15 converts to a thickness in lucite not water. Finally the compensator
thickness is
cXy= ATxy - PA(") (4.20)
where PA is the thickness of the preabsorber. PA is set to 1 mm less than the
minimum value of AT,,. This ensures the minimum compensator thickness is 1
mm rather than 0 mm leaving some material to support the thicker compensator
regions (Wagner 1982).
This simple compensation ignores two factors which should not be ignored;
multiple scattering and patient movement. To properly account for the first,
Monte Carlo studies in three dimensions must be made and this is unrealistic
if required for each patient. Such calculations have been done for studying the
general properties of introducing absorbers into the beam-line (Kanai et af 1991).
The second cannot be precisely accounted for as patient motion is almost always
unpredictable. Goitein (1978a) thus argues that the aim of the treatment plan
should be to predict the worst situation that can occur and ensure that even this
plan would be acceptable. Consider the two factors in tum:
To account for multiple scattering the maximum and the minimum path lengths
which a proton could follow are computed. These would give, respectively, the
minimum and maximum proton ranges, In three dimensions two surfaces would
result between which the actual proton range must lie. These path lengths can be
estimated by replacing in tum each CT voxel by first the smallest (to get minimum
path length) and second the largest (to get maximum path length) of itself and its
neighbours up to distances of about f 2 voxels. This procedure is not as good as a
full calculation but gives the range of uncertainty. Clinically, the bolus should be
adjusted so the most proximal of the two candidate 90% dose surfaces coincides
with the distal surface of the target volume. In his paper Goitein (1978a) writes
of lines rather than surfaces because only 2D CT data were available to him, but
recognized the need to make these computations in 3D, as expressed here.
Movement is similarly accounted for. The worst-case movement is estimated
and the elements of patient which could have intercepted the proton beam are
used to predict minimum and maximum ranges within which the actual range
must lie. For example, in the case of a high-density structure the beam must be
made additionally penetrating not only where the structure is but also where it
might be. Goitein (1978a) thus argues that single proton plans are not acceptable.
Plans should be made with worst-case calculations for movement and multiple-
scattering ensuring the required plan lies between the two extreme possible plans
(Chen 1983). At HCC-MGH the dose-planningcomputer is linked to an automatic
milling machine to make the compensators (Wagner 1982).
Urie et a1 (1984) present an altemative way to account for tissue movement.
The method is now the accepted technique at HCC-MGH, replacing the earlier
one above (Goitein 1978a, Urie et a1 1986a). A so-called 'simple bolus' is made
by the method described above using simple path lengths through the CT data.

No account need be taken of adjacent pixels to attempt to account for multiple

scattering or movement of tissues. This is done differently. Instead, once the
simple bolus has been computed, the thickness at any particular position in the
bolus is replaced by the minimum thickness at f d from that position, where d is 3-
5 mm. The resulting bolus is called the ‘expanded bolus’. The design philosophy
is that movement of the tissue by izd would now ‘not matter’ in the sense that
the beam would still be sufficiently penetrating to generate the required high-dose
volume. The worst that could now occur is that the dose might extend a little
beyond the target volume. There is no chance that any of the target volume would
be underdosed.
Urie et a1 (1 984) tested this extensively with phantoms worked up in the same
way as patients. They found that misregistration of a simple bolus with the
tissue led to very bad errors (more than 1 cm misjudgement of proton range).
However, when the expanded bolus was also deliberately misregistered, these
errors were reduced to at worst 2.5 mm in a phantom situation which was the
worst expected clinically. Of course if the expanded bolus and the phantom were
not misregistered, then the dosimetry was not quite as good as had the correctly
registered simple bolus been in use. This is the small price to pay for confidence in
accounting for tissue movement. It turns out that the finite size of the milling tool
(Wagner 1982) used for bolus construction actually itself introduces a measure of
‘expansion’.
Daftari et a1 (1991) have taken a different approach to assessing the importance
of random movement during charged-particle therapy. Treatment plans with
simulated patient movement were made; the dose-volume histograms were
constructed for organs at risk and converted to a measure of NTCP via the Lyman
model (see chapter 1). They concluded that random movements with maximum
size 2 mm, typical of clinical experience at the Lawrence Berkeley helium-ion
treatment facility, led to no appreciable increase in NTCP.
The compensating bolus should preferably be in contact with the patient but in
practice this may not be possible and an air gap will intervene (Urie et a1 1985,
Sisterson et a1 1987). Sisterson et a1 (1989) have analysed the effects of this air
gap on the location of the distal dose contours using Monte Carlo methods. Model
pairs of compensator plus phantom were created, as shown in figure 4.22. For the
simple bolus these pairs lock together with no air gap to give a uniformly thick
block. Several model pairs were considered in which the compensator was either
‘ridge first’, with an increased thickness to shorten the range, or ‘valley first’,
with a decreased thickness to lengthen the range. The valley and ridge could
take various apex angles and the air gap was variable. Protons were modelled
interacting with this arrangement and also with the ‘expanded bolus’ arrangement
shown in the lower part of the figure.
The behaviour was such that the amplitude of the perturbation on the proton
penetration increased with increasing air gap, with increasing pre-absorber
thickness (i.e. the uniform base of the bolus) and depended on the apex angle.
However, provided the air gap was less than 12 cm, the perturbation was never

'Expanded' boluses
lb) 'Ridge f i r s t ' aligned Id I Valley f i r s t * aligned

PROTONS PROTONS
i t t t t l i t 1 1
I I LUClTE 1
L w r e Removed
For "E~pomion"
1 LUCITE I LUCITE
Figure 4.22. The compensating bolus- ('patient' ) phantom arrangement

used to investigate the importance (or otherwise) of air gaps between the
bolus and the patient. The apex angle and air gap are defined in the figure.
( a ) 'Ridgelfirst' geometry for the simple bolus; the ridge harf simulates the
bolus and the valley halfthe patient; ( b ) 'ridge first' for the expanded bolus;
material is removed from the bolus to 'open it up'; (c) 'valleyfirst' geometry
for the simple bolus; the valley halfsimulates the bolus and the ridge halfthe
patient; ( d ) 'valleyfirst' for the expanded bolus; material is removedfrom the
bolus to 'open it up'. The direction of protons is shown by the arrows. {From
Sisterson et a1 (19891.)
more than 4 mm. Since the margin added at the planning stage to account
for uncertainties in penetration was typically 2 5 % of the maximum depth of
penetration (say 5% of 12 cm), the perturbation tums out to be less than this
uncertainty. It would therefore appear that air gaps are not a problem.
At the Japanese proton facility at Chiba, bolus construction also relies on CT
data but is much simpler and ignores tissue inhomogeneities, since the beam can
only penetrate 3.8 cm and is used mainly for irradiating the head and neck. At
Chiba a dental impression material, alginate, with a range correction factor of
1.06 relative to water is poured into little frames attached to the skin surface just

prior to treatment to create a flat proximal surface, Thin acrylic plates are used to
fine tune the compensation to account for changes in tumour volume during the
course of therapy (Akanuma et a1 1982).
The sharp penumbra is one of the compelling arguments for proton therapy.
However, introduction of scattering and beam-defining components into the beam
degrades the penumbra. The degrading components are the pre-absorber, the
beam-defining aperture, the compensating bolus and the effects of these depend
on the size of the drift (air) spaces. Characterizing all these effects is complex,
but Urie et a1 (1986b) have applied Monte Carlo methods to the problem. As one
might expect, penumbra increases with:
1. a decrease in proton residual range (increase in thickness of pre-absorber),
2. size of air gap between the beam aperture and the patient,
3. depth into the patient,
4. thickness of bolus.
Detailed curves were given by Urie et a1 (1986b). Particularly useful are a set
of ‘isopenumbra’ curves showing the combinations of circumstances which give
the same penumbra.
4.4.3. Theoretical treatment of thin inhomogeneities

In section 4.4.1 the influence of thin slivers of tissue inhomogeneity on proton
dose distributions was discussed. Depending on the width of the sliver the dose
distribution in the shadow of the inhomogeneity may get filled in by multiple
scatter. Conversely the main effect of a large slab of tissue inhomogeneity is to
alter the proton range. The effect requires compensation by external bolus.
Another concem is the dose-perturbing effect of thin inhomogeneities which
do not extend across the full cross section of the proton beam. These cause
local perturbations to the dose near, and in the shadow of, the lateral margin of
the inhomogeneity. Goitein (1 978b) has provided a very elegant mathematical
description of these effects, which now follows. Firstly, and qualitatively, the
effect of an edge discontinuity can be predicted without the use of mathematics.
Consider figure 4.23. A proton beam irradiates an edge of a thin inhomogeneity.
Particles in group I miss the edge and give rise to an unscattered step-like
fluence downstream (shown dotted). Particles in group I1 however intercept the
inhomogeneity, scatter, and give rise to a fluence falling from left to right in the
diagram. The resulting total fluence has the very characteristic shape shown as
the solid line. The fluence is exactly halved just below the edge and rises to 1.5
times the mean level just to the right of this. This is because the fluence near
Q, close to the edge, must be exactly half that at points such as Q’ because Q
lacks scatter from half the plane. On the other side of the edge-shadow at a
close point, such as P, the scatter is the same and adds to the fluence from the
group I particles to give an enhancement to 1.5. The total fluence profile is a
universal curve if the horizontal axis is considered to be the angular deviation
from the forward direction, since the scattering depends only on angular deviation

a' a 0 P P'
TRANSVERSE D/STANCE
RELATIVE ro EDGE
Figure 4.23. (Upper) shows schematically a parallel beam of protons

incident on a thin inhomogeneity only partly filling the field. (Lower) the
fluence close to the shadow of the edge. There is a universal curve (solid)
for the form of the fluence. being the sum of two components (dotted)from
the two groups of protons, shown as I and II. At the edge there is a sudden
discontinuity of size unity (see text for explanation). (From Goitein ( I 97861.)
of the measurement point from the forward direction (shown as 0 ) and so the
lateral extent of the perturbation increases with increasing distance beneath the
inhomogeneity. The above argument assumes the inhomogeneity is sufficiently
thin that particles do not loose much energy traversing it.
Goitein (1978b) derives the general formulae which give the edge perturbation
and can be applied to any geometry. Consider figure 4.24 in which a parallel beam
of protons irradiates a thin slab with aino protons per unit length. The factor ai
(1 for a uniform beam) can be non-unity to describe a non-uniform beam. The
index i labels the position in the scatterer; consider those photons incident on the
ith element. Multiple scattering gives rise to an angular distribution
f (e) de = [ 1/ai( 2 7 ~ ) ' /exp

~ ] (-e2/203 de (4.21)
where ai is the characteristic projected scattering angle which depends on the

atomic number of the material, the charge, velocity and momentum of the proton
and on the thickness and radiation length of the scatterer (Rossi (1956) formula).
The fluence of particles crossing a test region P of length dl at distance t from a
small element of length dx in the ith scatterer is
(ainodr) [ l/ai ( 2 ~ r ) ' /exp

~ ] (-02/2$) (dl/t) (4.22)
since
de = d l / t (4.23)

_ _ ---- - --
P
(6) 0, no
particles
per unit
4 1I--- ---I
11
length
dx
._ ..-. '.
.-.- \
--.-..
. .', \
I--
Figure 4.24. (Upper)showing a parallel beam of protons incident on a thin

scatterer, considered to be divided into elements labelled by i . The scattered
protons are shown figuratively with some reaching point P. (Lower) a single
scattering element i and an element d x within this, scattering to point P. The
totalfluence scattered to P from this element is the integration over the length
of the element. The totalfluence scattered to P from the whole scatterer is this
quantity integrated over all the elements i . (From Goitein (1978b)J
is the elemental angle subtended by dl at dx. In the absence of scatterer the fluence
at P for a uniform beam (ai = 1) would be simply nodi. The ratio Fi of the fluence
with scatterer to the fluence without scatterer is the result of taking the ratio of the
quantity in equation (4.22) to this value and integrating over the range of x. This
ratio is
Fi = (l/nodl) (ainodx) [ l / q ( 2 x ) ' / * ]exp (-0'/2$) (dI/t) . (4.24)
If we make a small-angle approximation that 0 = (I - x ) / r so that de = - d x / t ,

this reduces to
Fi = ai/ [ l / q ( 2 x ) ' / ' ] 1; exp ( - 0 ' / 2 4 de. (4.25)
The fluence at P from all elements labelled by i is then the sum

F='YF~ (4.26)

This can be translated to dose D by a multiplicative constant, the dose-to-fluence

ratio at P in the absence of inhomogeneity
D = C d j Fj. (4.27)
1
Equations (4.25) and (4.26) involve only the specification of the homogeneity
of the beam (through ai) and a knowledge of the characteristic scattering angle
(through ai). Everything else is a geometrical integration.
The above analysis holds provided the following (very reasonable set of)
approximations hold:
1. the material is thin and does not significantly affect proton range,
2. the inhomogeneity of finite thickness is ‘condensed’ to be considered
as an infinitely thin plane for mathematical argument with the same scattering
properties,
3. the discontinuity is infinite in the third dimension (realistic because
perturbing effects are local to the edge),
4. the projected scattering angle can be represented by a Gaussian distribution,
5 . the scattering is sufficiently forward peaked that one can make the small-
angle approximation (see above equation (4.25)) that f3 2: tan (e)
6. the particles are monoenergetic.
Fan-beam geometry can be incorporated easily into this model. Equation (4.25)
requires specification of the two limits of integration; the largest and smallest
scattering angles. These can be specified equally well in fan-beam geometry
(figure 4.25). Goitein (1978) also shows that the effect of tissue inhomogeneities
both upstream and downstream from the inhomogeneity in question can be
similarly accounted for since they alter the distribution of the angles of incident
protons.
Retuming to equation (4.25), if the thin inhomogeneity extends for half the
proton field, is set to n/2. For a uniformly irradiating proton beam, the
calculations are easy to do by hand because equation (4.25) reduces to looking
up values of the error function erf ( x ) . Under other circumstances a computer is
needed to evaluate equation (4.25). The formula was applied to tissue-air and
tissue-bone interfaces. Whereas the former introduce 50% perturbations close to
the edge, it tums out the latter only introduce a maximum of 9% variations. Thus
tissue-air boundaries may be much more important in proton therapy than tissue-
bone boundaries as far as edge-scatteringeffects are concerned. Even these values
are reduced if the beam is divergent and there is scattering tissue above and below
the inhomogeneity (as of course is always the case in clinical reality). This avoids
what might otherwise be a disturbing and serious clinical problem.
Goitein et al (1978) pursued this further, calculating and measuring the
perturbations from a variety of different geometrical interfaces for proton (and
incidentally electron) beams. For all situations considered there was excellent
agreement between theory and experiment.

5OLKCE
Figure 4.25. Thin-inhomogeneity scattering of a divergent beam showing

the definition of the range of angles in the integration of scatterfrom a small
elemental length of scatterer, labelled by i to the point P. (From Goitein
(1978).)
It is apparent that the formalism presented in this section cannot become the
basis of a treatment-planning tool. It would be just too complicated to properly
treat the effects of all the body inhomogeneities. Indeed, as Goitein et a1 (1978)
showed, tissue inhomogeneities close to each other produce fluence interference
effects. From a computational viewpoint there are too many interactive decisions
needed in working out equation (4.25) for complex geometries. Instead the
formalism and the results therefrom should be used to estimate the effects
of inhomogeneities in perturbing the distribution of dose in specific clinical
situations.
4.4.4. The dose-perturbing effect of thick inhomogeneities

Whilst an analytic technique has been developed for calculating the dose-
perturbing effects of thin inhomogeneities, the perturbation from thick inhomo-
geneitie9 cannot be represented analytically. ‘Thick’ means ‘of comparable di-
mension to the proton range in the material’. This is because some of the as-
sumptions which went into the ‘thin’ model are violated by the ‘thick’ model.
For example the ‘condensed’ material approximation cannot be made and the mo-
noenergetic scattering-angle approximation is invalid. The small angle ‘paraxial’
approximation still holds.
The answer lies with Monte Carlo techniques. Goitein and Sisterson (1978)
have numerically modelled the scattering in two-dimensions from slivers of bone
(or teflon substitute) in tissue, when the slivers were comparable to or longer
than the proton range. The scattering medium is pixellated and then the Monte
Carlo is a two-step process. Firstly, the protons are transported small distances in
straight lines with corresponding energy losses and range changes. Secondly, the
proton’s direction is altered by randomly selecting from the well known Gaussian
distribution with standard deviation given by the Rossi formula (see section 4.4.3).

Some 0.1 M histories were followed and flux and dose were plotted as a histogram.
A typical result from this work has already been discussed (figure 4.18). They
modelled a realistic beam, range-modulated to give a uniform dose over a depth
of 16 cm, characteristic of what could be achieved with the HCC-MGH proton
facility. The following results, for slivers spanning the central axis of the beam as
in figure 4.18, are abstracted from the large volume of Monte Carlo data created:
1. wide slivers reduce the penetration from the value in homogeneous water,
2. as the slivers become narrower, central-axis dose builds up beyond the
theoretical (wide sliver) range because of ‘in-scatter’ from the surrounding water,
3. dose is enhanced laterally to thin slivers in the surrounding water,
4. if additional angular smearing is added to an otherwise parallel beam, the
dose-perturbing effects are considerably reduced,
5. rocking the inhomogeneity (equivalent to a positioning error) causes
similar reductions in dose perturbation,
6. irradiation of a sliver by a parallel beam not parallel to the long edge of the
sliver, dramatically reduces the dose perturbation.
Unfortunately, multiple and complex tissue inhomogeneities can have a totally
destructive effect on the Bragg peak and Urie et a1 (1986a) have carried out
experiments with protons (as well as other heavy ions) to show this. There are
circumstances in which multiple scattering cannot be disregarded and cannot be
simply accounted for by the methods of bolus construction discussed in section
4.4.2. Urie et af (1 986a) argue that nothing short of full Monte Carlo computations
will really suffice. One must recall that the construction of compensators can
only account for the changes in ‘areal density’, i.e. the product of density
and path length, together with methods of attempting simple accounting for
movement and multiple scatter. The methods are not assumption-free; there are
classes of inhomogeneities which can ‘fool’ the algorithm for compensator bolus
construction. This problem remains the ‘Achilles’ heel’ of proton therapy.
4.4.5. The differentialpencil-beam model for proton dose calculation

In recognition of the difficulties of properly accounting for multiple scattering,
Petti (1992) has developed a differential pencil-beam model for protons. The
philosophy is the same as underlies the differential pencil-beam formalism for
photons, namely the important physics is built into the elemental dose distribution
and then complex field calculations are made by integrating the elemental
distribution in the appropriate geometry (cf section 2.5.2.1). The result is much
more accurate than simple ray-tracing, but is far less time consuming than full
Monte Carlo calculations. Petti (1992) validated the following formalism by
showing that, for five sample geometries, the predicted dose distributions were
very close to the results of full Monte Carlo calculation and differed from the
results of simple ray-tracing in many important respects. The differences became
more pronounced as the complexity of the problems increased. The formalism
accounts for tissue inhomogeneities in the path of the protons.

Proton treatment planning 21 1
Figure 4.26. The geometry for defining the differential pencil-beam dose
distribution for protons. The beam is of infinitessimal area and directed down
the z axis. The DPB dose distribution, F (r, z ) , is the dose deposited in the
+ +
annulus shown of radius r to r d r and depth z to z d z , normalized to the
dose Do in the Bragg peak or SOBP. (From Petti ( I 992).)
Define the differential pencil-beam dose distribution, F ( r , z), as the dose

deposited in a homogeneous water phantom in an annular ring between radii r and
+ +
r d r and depth z and z dz, normalized to the maximum dose DOin the Bragg
peak (or SOBP) of the beam (figure 4.26). The pencil beam has infinitessimal
area but contributes dose to a finite range of radii because of multiple scattering.
F (r, z ) may be computed by Monte Carlo methods. The dose deposited at any
point ( r , z ) per unit beam area is Do F (r, z) /2nr. Suppose the proton intensity
(protons per unit area) at any position in the beam is @ (r, e), where (r, 0) is a
general point, with origin at the point of dose calculation, in the x y plane normal
to the incident direction of the protons. The dose at any Cartesian point ( x , y , z )
is then
In equation (4.28) r is the radius from the centre of the contributing pencil beam
at ( x ' , y ' ) to the calculation point ( x , y) at depth z (i.e. ( x - x ' ) = r cos (e) and
( y - y') = r sin (e)). Equation (4.28) simply expresses a summation of the dose
from the set of elemental pencil beams appropriately weighted for intensity. The
upper limit on the radial integral represents the maximum radial distance for which
multiply scattered protons can contribute to the dose at point ( x , y , z ) . In practice
this is fairly small for protons leading to short computational times. If the proton
intensity is uniform, equation (4.28) reduces to
Inhomogeneities are accounted for by replacing z by the water-equivalent

pathlength z , to obtain
D ( x , y , z ) = DOlrw
1'" (I'
21T
(r' e) dr de (4.30)

2 12 The physics of proton radiotherapy
where
zw = 1’p (r,8, z’) dz’ (4.3 1)
and p is the electron density of tissues. The integration in equation (4.31) is

performed along the ray-path of the protons and, if desired, can account for beam
divergence. It does not account for the position of the tissue inhomogeneity down
the ray-path.
Petti (1992) also developed a full Monte Carlo code modelling proton multiple
scatter, using it to predict benchmark dose distributions for five model geometries.
She then compared the results with predictions from the pencil-beam model
with very good agreement. The ray-tracing algorithm which ignores multiple
scatter was adequate for homogeneous phantoms but became increasingly more
inaccurate by comparison with the gold standard as the complexity increased.
Unlike photon-beam calculations involving pencil beams, the DPB formulation
for protons is not enormously slower than simple ray-tracing. Petti (1992) quotes
a reduction in speed of about a factor 12. If tissue inhomogeneities have a
complicated geometry, the effort is most worthwhile because ray-tracing can give
quite erroneous results.
4.5. SUMMARY
Radiotherapy with protons is nearly four decades old but, possibly because its
birth coincided with the development of megavoltage photon therapy and in view
of the enormous expense and consequent rarity of proton facilities, has not become
widely available. However, it has been persued vigorously in a number of centres
and now is poised to capture greater interest as more purpose-built centres begin
to come on stream.
The inherent physical properties of protons interacting with tissue make them
ideal candidates for conformal therapy. The Bragg peak can be spread out by
range-modulating wheels and the physical area of the beam can be enlarged from
the elementary beam by double-foil-plus-annuli scattering or by spot scanning.
Proton treatment planning is very complex and needs to account for tissue
inhomogeneities and the possibility of patient movement. Because of the rapid
decline of dose distal to the SOBP,patient movement is potentially more dangerous
for proton than photon irradiation because severe underdosage could arise unless
the measures described are put into practice. Patient fixation is important.
Many aspects of proton dose distributions can only be studied in model
situations because of the complex interaction physics. Monte Carlo studies
have been performed to assess the importance of disturbing features such as
inhomogeneities and movement.

References 2 13
REFERENCES
Akanuma A, Majima H, Furukawa S , Okamoto R, Kutsutani Y, Nakamura,

Tsunemoto H, Morita S , Arai T, Kurisu A, Hiraoka T, Kawachi K and Kanai T
1982 Compensation techniques in NIRS proton beam radiotherapy Int. J . Rad.
Ambrosimov N K, Vorobev A A, Zherbin E A and Konnov B A 1985 Proton
therapy at the cyclotron at Gatchina, USSR Proc. Acad. Sci. USSR 5 84-91
Antoine F S 1987 Proton beam therapy4ost versus benefit J . Nut. Cancer Inst.
81 559-562
Antoine F S 1991 International collaboration may enhance proton therapy research
J . Nut. Cancer Inst. 83 9-10
Awschalom M W, Cole F T, Lidahl P V, Mills F E and Teng L C 1987 Design of
a proton therapy accelerator Med. Phys. 14 468
Beeckman W, Jongen Y, Laisne A and Lannoye G 1991 Preliminary design of a
reduced cost proton therapy facility using a compact, high field isosynchronous
cyclotron Nucl. Instrum. Methods in Phys. Res. B56J57 1201-1204
Bichsel H 1972 Passage of charged particles through matter American Inst of
Physics Handbook section 8 (New York: McGraw-Hill) pp 142-189
Blattman H (ed) 1986 Proton therapy of choroidal melanomas (OPTIS) Swiss
Institute for Nuclear Research Medical Newsletter 8 32
Blosser H G 1989 Compact superconducting synchrocyclotron systems for proton
therapy Nucl. Instrum. Methods in Phys. Res. B40J41 1326-1330
Bonnett D E 1991 The role of protons in radiotherapy Proc. 1990 Int. Con$ on
Applications of Nuclear Techniques ed G Vourvopoulous and T Paradellus
(Singapore: World Scientific) pp 109-130
Bragg W H and Kleeman R 1904 On the ionisation curves of radium Phil. Mag.
8 726-738
Chen G T Y 1983 Computed tomography in high LET radiotherapy treatment
planning Computed tomography in radiation therapy ed C C Ling, C C Rogers
and R J Morton (New York: Raven) pp 221-227
Chen G T Y, Singh R P, Castro J R, Lyman J T and Quivey J M 1979 Treatment
planning for heavy ion radiotherapy Int. J. Rad. Oncol.B i d . Phys. 5 1809-1819
Chen G T Y and Goitein M 1983 Treatment planning for heavy charged particle
beams Advances in Radiation Therapy Treatment Planning ed A E Wright and
A L Boyer (New York: American Institute of Physics) pp 514-541
Chu W T, Renner T R and Ludewigt B A 1989 Dynamic beam delivery for three-
dimensional conformal therapy Proc. Eulima Workshop on the potential value
of light ion beam therapy (Nice, 1988) ed Chauvel P and Wambersie A
Chuvilo I V, Goldin L L, Khoroshkov V S , Blokhin S E, Breyev V M, Vorontsov
I A, Ermolayev V V, Kleinbock L, Lomakin M I, Lomanov M F, Medved’ Ya
V, Miliokhin N A, Narinsky V M, Pavlonsky L M, Shimchuck G G, Ruderman
A I, Monzul G D, Shuvalov E L, Kiseliova V N, Marova E I, Kirpatovskaya L
E, Minakova E I, Krymsky V A, Brovkina A K, Zarubey G D, Reshetnikova I

2 14 The physics of proton radiotherapy
M and Kaplina A V 1984 ITEP synchrotron proton beam in radiotherapy Int.

J . Rad. Oncol. B i d . Phys. 10 185-195
Cole D J 1990 The case for proton therapy in the UK private communication Io
AEA September 1990
Coutrakon G, Bauman M, Lesyna D, Miller D, Nusbaum J, Slater J, Johanning J,
DeLuca P M Jr, Siebers J and Ludewigt B 1991a A prototype beam delivery
system for the proton medical accelerator at Loma Linda Med. Phys. 18 1093-
1099
Coutrakon G, Miller D, Kross B J, Anderson D F, DeLuca P Jr and Siebers J 1991b
A beam intensity monitor for the Loma Linda cancer therapy proton accelerator
Med. Phys. 18 (4) 8 17-820
Daftari I, Petti P L, Collier J M, Castro J R and Pitluck S 1991 The effect of patient
motion on dose uncertainty in charged particle irradiation for lesions encircling
the brain stem or spinal chord Med. Phys. 18 1105-1 115
Goitein M 1972 Three dimensional density reconstruction from a series of two
dimensional projections Nucl. Instrum. Methods 101 509-5 18
- 1977 The measurement of tissue heterodensity to guide charged particle
radiotherapy Int. J . Rad. Oncol. Biol. Phys. 3 27-33
-1978a Compensation for inhomogeneities in charged particle radiotherapy
using computed tomography Int. J . Rad. Oncol. B i d . Phys. 4 499-508
-1978b A technique for calculating the influence of thin inhomogeneities on
charged particle beams Med. Phys. 5 (4) 258-264
- 1982 Applications of CT in radiotherapy treatment planning Progress in
medical radiation physics ed C G Orton (New York: Plenum) pp 195-293
Goitein M, Abrams M, Gentry R, Urie M, Verhey L and Wagner M 1982 Planning
treatment with heavy charged particles Int. J . Rad. Oncol. Biol. Phys. 8 2065-
2070
Goitein M, Chen G T Y, Schneider R J and Sisterson J M 1978 Measurements and
calculations of the influence of thin inhomogeneities on charged particle beams
Med. Phys. 5 (4) 265-273
Goitein M and Munzenrider J 1984 Proton therapy: good news and big challenges
Int. J. Rad. Oncol. Biol. Phys. 10 319-320
Goitein M and Sisterson J M 1978 The influence of thick inhomogeneities on
charged particle beams Rad. Res. 74 217-230
Goldin L L and Monastyrsky 1978 Analytical calculation of ridge filter
configuration Medic. Radiol. 23 (12) 65-68 (in Russian)
Gottschalk B 1987 Design of a hospital-based accelerator for proton radiation
therapy Nucl. Instrum. Methods in Phys. Res. B24/25 1092-1095
Graffman S,Brahme A and Larsson B 1985 Proton radiotherapy with the Uppsala
cyclotron. Experience and plans Strahlentherapie 161 764-770
Janni J F 1982 Proton range-energy tables, 1 keV-I0 GeV Atomic Data and
Nuclear Tables 27 147-339
Jongen Y 1992a private communication
- 1992b An improved isocentric gantry with reduced diameter for proton

References 2 15
therapy by scanned beam IBA paper Chemins du Cyclotron,2 b- 1348 Louvain-

la-Neuve, Belgium
Jongen Y, Beeckman W and LaisnC A 1992 Development of a low-cost compact
cyclotron system for proton therapy IBA paper Chemins du Cyclotron,2 b-1348
Louvain-la-Neuve, Belgium
Kanai T, Kawachi K and Hiraoka T 1991 An irradiation facility and a beam
simulation program for proton radiation therapy Nucl. Instrum. Methods in
Phys. Res. A302, 158-164
Kanai T, Kawachi K, Kumamoto Y, Ogawa H, Yamada T and Matsuzawa H 1980
Spot scanning system for radiotherapy Med. Phys. 7 (4) 365-369
Kantor G, Destembert B, Breteau N and Schlienger M 1985 Les protons en
radiothtrapy’ Gazette Mtdicale 92 101-103
Kawachi K, Kanai T, Matsuzawa H and Inada T 1983 Three dimensional spot
beam scanning method for proton conformation radiation therapy Acta Radiol.
Supp 364
Khoroshkov V S , Barabash L Z, Barkhudaryan A V, Goldin L L, Lomanov M F,
Onosovsky K K and Plyashkevich L N 1969 Proton beam of ITEP accelerator
for radium therapy Med. Radiol-Mosk. 14 58-62 (in Russian)
Kitigawa T 1988 Proton beam therapy J . Rad. Res. 29 6
Koehler A M and Preston W M 1972 Protons in radiation therapy Radiology 104
191-1 95
Koehler A M, Schneider R J and Sisterson J M 1975Range modulators for protons
and heavy ions Nucl. Instrum. Methods in Phys. Res. 131 437-440
-1977 Flattening of proton dose distributions for large field radiotherapy Med.
Phys. 4 297-301
Larsson B, Leksell Land Rexed B 1963The use of high energy protons for cerebral
surgery in man Acta Chirurg. Scand. 125 1-7
Larsson B, Leksell L, Rexed B, Sourander P, Mair W and Andersson B 1958 The
high-energy proton beam as a neurosurgical tool Nature 182 1222-1223
Larsson B, Liden K and Sarby B 1974 Irradiation of small structures through the
intact skull Acta Radial. Ther. Phys. B i d . 13 512-534
Lennox A J 1991 Hospital-based proton linear accelerator for particle therapy and
radioisotope production Nucl. Instrum. Methods in Phys. Res. B56/57 1197-
1200
Miller D W, Slater J M, Awschalom M W and Allen M A 1987 Design of a
hospital-based proton beam therapy centre Med. Phys. 14 468
Montelius A, Grusell E, Russell K, Blomquist E, Pellettieri L and Lilja A
1991 Proton irradiation of arteriovenous malformations in the brain using a
positioning system with titanium markers Proc. 1st biennial ESTRO meeting
on physics in clinical radiotherapy (Budapest, 1991) p 31
Munzenrider J E, Austin-SeymourM, Blitzer P J, Gentry R, Goitein M, Gragoudas
E S , Johnson K, Koehler A M, McNulty P, Moulton G, Osbome E, Seddon J M,
Suit H D, Urie M, Verhey L J and Wagner M 1985 Proton therapy at Harvard
Strahlentherapie 161 75&763

Petti P 1992 Differential-pencil-beam dose calculations for charged particles Med.

Phys. 19 (1) 137-149
Phillips M H, Pedroni E, Blattmann H, Boehringer T, Coray A and Scheib S
1992 Effects of respiratory motion on dose uniformity with a charged particle
scanning method Phys. Med. Biol. 37 223-234
Potts T M, Miller D, Prechter R, Prichard B and Slater J M 1988 The conceptual
design of a patient handling system for isocentric proton beam therapy Med.
Phys. 15 798
Raju M R 1980 Heavy particle radiotherapy (New York: Academic) pp 188-25 1
Rossi B 1956 High energy particles (Englewood Cliffs: Prentice Hall) 2nd edn
P 64
Sharma J, Bedi B S and Mukhopadhyay S 1991 Linear accelerator for cancer
therapy-an indigenous effort Electronics Information and Planning 18 167-
177
Sisterson J M 1990 Overview of proton beam applications in therapy Nucl.
Instrum. Methods in Phys. Res. B45 718-723
- (ed) 1991 The proton therapy program at Louvain la Neuve Particles
Newsletter no 8, June 1991
-(ed) 1992 Particles Newsletter no 9, Jan 1992
Sisterson J M, Cascio E, Koehler A M and Johnson K N 1991Proton beam therapy:
reliability of the synchrocyclotron at the Harvard Cyclotron Laboratory Phys.
Med. B i d . 36 285-290
Sisterson J M and Johnson K N 1985 As we approach 3000: proton radiation
therapy at the Harvard Cyclotron Laboratory Med. Phys. 12 546
- 1986 As we approach 3000: proton radiation therapy at the Harvard
Cyclotron Laboratory AAMD Journal 11 21-30
-1987 Proton radiation therapy: the Harvard cyclotron experience Med. Phys.
14 469
Sisterson J M, Urie M M and Koehler A M 1987 Factors affecting the precise
shaping of range in proton beams Med. Phys. 14 468
Sisterson J M, Urie M M, Koehler A M and Goitein M 1989 Distal penetration of
proton beams: the effects of air gaps between compensating bolus and patient
Phys. Med. Biol. 34 1309-1315
Slater J M, Archambeau J 0, Miller D W, Notarus M I, Preston W and Slater J
D 1992 The proton treatment center at Loma Linda university medical center:
rationale for and description of its development Int. J . Rad. Oncol. Biol. Phys.
22 383-389
Slater J M, Miller D W and Archambeau J 0 1988 Development of a hospital-
based proton beam treatment centre Int. J. Rad. Oncol. B i d . Phys. 14 761-775
Slater J M, Slater J D, Archambeau J 0 and Miller D 1990 Proton therapy: a
modality for expanding the parameters of precision radiation therapy Loma
Linda Radiology Medical Group Inc: Rad. Sci. report 2 (1) Jan 1990
Suit H D, Goitein M, Munzenrider J, Verhey L, Urie M, Gragoudas E, Koehler A,
Gottschalk B, Sisterson J, Tatsuzaki H and Miralbell R 1990) ‘Increased

References 217
efficacy of radiation therapy by use of proton beam Strahlentherapie und

Onkologie 166 (1) 40-44
Suit H D, Goitein M, Tepper J, Koehler A M, Schmidt R A and Schneider R 1975
Exploratory study of proton radiation therapy using large field techniques and
fractionated dose schedules Cancer 35 1646-1 657
Suit H D and Verhey L J 1988 Precision in radiotherapy Brit. J . Radiol. (Suppl22)
pp 17-24
Tobias C A, Lawrence J H, Lyman J, Born J L, Gottschalk A, Linfoot J, McDonald
J 1964 Progress report on pituitary irradiation Response of the nervous system
to ionising irradiation ed T J Haley and R S Snider (New York: Little, Brown
and CO)
Tobias C A, Roberts J E, Lawrence J H, Low-Beer B V A, Anger H 0, Bom J L,
McCombs R, Huggins C 1956 Irradiation hypophysectomy and related studies
using 340 MeV protons and 190 MeV deuterons Peaceful Uses of Atomic
Energy 10 95-96
Tsunemoto H 1988 Radiation therapy with accelerated particles J . Rad. Res. 29 6
Tsunemoto H, Morita S , Ishikawa T, Furukawa S , Kawachi K, Kanai T, Ohara
H, Kitagawa T and Inada T 1985 Proton therapy in Japan Rad. Res. 8 (Suppl)
S235-S243
Urie M, Goitein M, Holley W R and Chen G T Y 1986a Degradation of the Bragg
peak due to inhomogeneities Phys. Med. B i d . 31 1-15
Urie M, Goitein M, Sisterson J M and Koehler A M 1985 Proton dose distribution
influenced by changes in drift space between patient and beam modifying
devices Med. Phys. 12 546
Urie M, Goitein M and Wagner M 1984 Compensating for heterogeneities in
proton radiation therapy Phys. Med. Biol. 29 553-566
Urie M, Sisterson J M, Koehler A M, Goitein M and Zoesman J 1986b Proton
beam penumbra: effects of separation between patient and beam modifying
devices Med. Phys. 13 (5) 734-740
Verhey L J and Munzenrider J E 1982 Proton beam therapy Ann. Rev. Biophys.
Bioeng. 11 331-357
Wagner M S 1982 Automated range compensation for proton therapy Med. Phys.
9 ( 5 ) 749-752
Webb S 1989 Optimisation of conformal radiotherapy dose distributions by
simulated annealing Phys. Med. Biol. 34 1349-1369
Wilson R R 1946 Radiological use of fast protons Radiology 47 487-491

CHAPTER 5
CONFORMAL RADIOTHERAPY WITH

A MULTILEAF COLLIMATOR
5.1. INTRODUCTION
Multileaf collimators (MLC) are regarded presently as the state-of-the-art method

for generating arbitrary (and generally irregularly) shaped fields for x-ray therapy.
They are still research tools in a number of specialized centres interested in
conformation radiotherapy and are not widely available. They are also used for
shaping neutron beams in view of the ease of varying the field shapes compared
with moving very heavy volumes of ordinary block collimators (Eenmaa et a1
1985, Chu and Bloch 1986,1987, Brahme 1988, Wambersie 1990, Bewley 1989).
This chapter reviews the technology. Planning aspects of using a multileaf
collimator, including the exciting possibility of tailoring the intensity of the
radiation spatially across the field, are discussed in chapter 2. The multileaf
collimator is sufficiently important to conformal radiotherapy to justify separate
consideration here from other features of treatment machines which are discussed
in chapter 7. The idea of moving leaves for radiation collimation is in fact rather
an old one (Robinsohn 1906).
5.1.1. Early history

Two patents for means of collimating radiation may be seen as landmarks of some
importance. The first (McGunnigle 1926/29) described apparatus for collimating
not an ionizing radiation but an optical field. It is worthy of attention in that it
describes how to collimate an irregular shape. The second (Green and McColm
1952/54) was a method for collimating ionizing radiation to square or rectangular
fields, but with a novel approach which bears on MLCs.
The essence of McGunnigle’s collimator may be grasped from figure 5.1. Four
plates, shown as ‘9’, are able to move relative to a groundplate ‘ 5 ’ , which is set up
such that the direction of radiation is at right angles to it. Two of the plates ‘9’ are
on one side of the groundplate and two (shown dotted) on the other. Each plate is
equipped with a pin ‘9’ which is able to slide and rotate in slots ‘7’. Each plate
218
Zntroduction 219
5-i 2 1
Figure 5.1. McGunnigleS collimator for irregular fields. Four collimating

plates ‘8’ can be translated and rotated by pairs of screws to define an
aperture with straight sides, but not necessarily rectangular. The application
was not to collimate radiation fields, but apparatus of this type could be made
to perform this task. (From McGunnigle (1924-26).)
also carries two traveller posts ‘22’ which are able to rotate relative to the plates.
The groundplate is equipped with two similar posts ‘17’ for each moving plate.
Threaded rods ‘16’ connect the posts ‘22’ to the other posts ‘17*,the unthreaded
parts of the rods being held firmly at ‘17’ so the rods can rotate but not slide.
Conversely, the threaded parts of the rods pass through the posts ‘22’ so that as
the rods rotate the posts are moved along the length of the rods.
It should now be clear that by tuming the two rods controlling each plate, the
plate may be moved. If both screws turn the same way, the plate moves bodily
forwards or backwards; if one screw is tumed one way and the other tumed the
other way, then the plate can be made to rotate. Each of the four plates is controlled
independently in the same way.
McGunnigle intended his apparatus for collimating projected light to specific
shapes (for example illuminating a painting without the frame). If the plates were
of appropriate thickness and material one could imagine a collimator for ionizing
radiation built along the same lines. A mechanism similar to this idea has in fact
been used very recently for a collimator for stereotactic radiosurgery (see chapter
3) (Leavitt et a1 1991).
Green and McColm ( 1952/54) developed a collimator for a cobalt or therapeutic
x-ray unit in which four lead blocks were able to move to create either a square
or rectangular field. What was novel about this development was that the four
blocks all lay in the same plane with the inside face of each one abutting the end
face of its nearest neighbour in a clockwise direction (figure 5.2). As each block
moved it carried with it that nearest neighbour, which clearly would then simply
slide along the end face of the opposing block. The abutting faces were provided
with tongue and grooving to eliminate stray radiation between the blocks. The
confining of the four blocks to a single plane was important. This is essentially

220 Conformal radiotherapy with a multileaf collimator
Figure 5.2. Showing the arrangement used by Green and McColm for
coplanar movement of four collimator blocks. (see text for description).
(From Green and McColm (1952-541.)
the same feature seen in MLCs; of course Green and McColm’s apparatus could
not provide for irregular fields.
Robinsohn (1906) patented an attenuating screen comprising multiple leaves to
define an irregular field for a diagnostic x-ray set. The diagram of the apparatus
(figure 5.3) does not show any method of automating the movement of the
leaves. The idea resurfaced (figure 5.4) with the German patent by Maas and
Alexandrescu ( 1982).
The idea of the multileaf collimator for radiation treatment machines appears to
have originated with Gscheidlen (1 959), who patented a device in which four sets
of orthogonally disposed leaves could be moved to create an irregular field shape.
The device is shown in figures 5.5 and 5.6 where, for simplicity, an arrangement
with just five leaves per set is shown. The central leaf appears thicker than the
other four in the set.
The leaves ‘10’ were positioned upright in the direction of the central ray
of the beam and the beam was laterally limited by the edges or walls of the
upright collimator leaves. The leaves were disposed so as to move in the plane
perpendicular to the central ray and supported by the plane surface ‘9’. The leaves
were guided by the angle bars ‘12-15’, and prism-shaped blocks ‘12a-15a’ and
screws ‘ 16, 17’ were provided whereby the pressure on the leaves could be varied.
The leaves were moved by push rods ‘19’ actuated by cranks ‘18’, which in
turn were driven by friction disks ‘21’ carried by a shaft ‘20’ by means of grooves
and tongues and pressed against the cranks ‘18’ by springs ‘23’. By tuming the
knob ‘26’ the movement was actuated. The final resting position of each leaf was
determined by a setting template ‘28’ in which pins ‘30’ could be positioned. The
leaves would come to rest at these positions (which in general were different for
each leaf) after which any further tuming of the screw ‘26’ would simply result in
slipping at the friction rings. By means of endless pulleys ‘27’, opposing pairs of
leaves were simultaneously adjusted. In this way the shape of the irregular field

Introduction 221
ILl0 /I
/'
I:I I 1
Figure 5.3. The multi-element collimatorfor diagnostic radiology, patented

by Robinsohn. The leaves are shown at '23' and were positioned by hand to
give an irregularly shapedjield ' 2 ' . (From Robinsohn (19061.)
Figure 5.4. Showing the collimator patented by Maas and Alexandrescu.

This was more advanced than that of Robinsohn, in that it had the mechanism
shown for moving the leaves '26'. Once again the application was in
diagnostic radiology. (From Maas and Alexandrescu ( I 9821.)
was only limited by the quantization introduced by the width of the leaves and
separation between the pins '30'.
The patent showed variations on this theme whereby several collimator systems

Figure 5.5. A schematic view of Gscheidlen’s multileaf collimator patented

in 1959, showing just two of thefour sets of leaves. The actuating mechanism
and the template for setting the jield shape may be seen, as well as some of
the guide bars (see text for description). (From Gscheidlen (19591.1
b
,.,%%Q 25
I’ 21
Figure 5.6. A plan view of Gscheidlen’smultileaf collimator and, alongside,

the details of the actuating mechanism for moving the leaves (see text for
description). (From Gscheidlen (19-79),)
of this type might be arranged one above the other to provide greater collimation,
and whereby arcuate leaves moving in arcuate guides (and multiple versions
thereof) could be constructed to limit the field in such a way that the edge of the
field lay tangential to the edges of the leaves.
These collimators were shown attached to a%o treatment head. They embody
most of the ideas germaine to modem multileaf collimators with motor drives.
However, the use of orthogonal pairs has not been carried over to the modem
multileaf collimator. This orthogonal arrangement is possible in a single plane
(unlike the conventional pairs of single lead-block collimators, one pair of which
has to reside above the other), because there is no possibility of collisions between
the orthogonally disposed collimators, since they comprise multiple leaves. This
collimator does not appear to have been developed commercially, and indeed was
technology ahead of its time.

Modern developments 223
5.2. MODERN DEVELOPMENTS
The concept of multileaf collimation is an attractive one since it dispenses with

the cumbersome business of casting and positioning individual blocks. Under
computer control the setting of arbitrary shapes can be very fast. There can be
a high degree of conformity with the beam’s-eye-view of the treatment volume
with reduced irradiation of normal tissue. The shapes can be fine-tuned, and even
varied during the course of fractionated delivery. The arrangement lends itself to
image-driven treatment planning. A multileaf collimator opens up the possibility
of using certain beam orientations which would not be useable with blocked fields
because of collisions between the blocking tray and the patient. This is one of the
exciting new possibilities with the multileaf collimator (Williams 1992).
Multileaf collimators can be characterized by:
1. number of leaf pairs,
2. width of leaves (or projected width at the isocentre),
3. compatability with accelerator type,
4. maximum field size,
5 . the ‘over-run distance’ (see section 5.2.2).
A number of collimators are catalogued this way in table 5.1.
5.2.1. BrahmeS patent

Brahme (1984, 1985/87, 1987, 1988) developed an MLC with 32 pairs of tungsten
leaves which moved on a circular orbit. This so-called ‘double-focused MLC’
was characterized by a very small (5 mm) penumbra, comparable to that obtained
with diverging beam blocks of low melting-point alloys. The high performance
was also due to the helium atmosphere inside the head, reducing multiple electron
scatter and the production of Compton electrons. The collimator accomodated
x-rays between 2.5 and 50 MeV and also 50 MeV electrons. The projected leaf
width at a 1 m isocentre was 1.5 cm, but the ‘staircase’ irregularity in the field
shape due to the finite width of the leaves was smoothed in the resulting isodose
contours. Similar observations were made for another collimator by Boyer et a1
(1991).
Brahme’s (1985/87) patent contains details of a number of features which are
desirable for a multileaf collimator and which improve on the rather simpler
collimator patented by Gscheidlen (1959). These include:
0 The leaves were curved such that both their upper and lower surfaces form part
of an imagined circle having its centre at the source of radiation.
0 The leaves were tapered in cross section and the group of 32 pairs, when viewed
end on, have a fan-like appearance again centred at the source. This ensures
that, whatever the positioning of the leaves, the tangents to all the edge surfaces
converge at the source, ensuring a minimal penumbra. These two features were
termed ‘double focusing’ (figure 5.7).

Figure 5.7. Detailed schematic of the arrangement of the leaves in the

double-focused mode. This is the ‘end-on’ view of the apparatus shown in
figure 5.8. The leaves are focused to the source. ‘50’ are the ends of the
motors driving the leaves ‘46’.(From Brahme (1 985187j.)
The leaves were supported on roller bearings (shown as ‘36’, ‘37’ and ‘38’ in
figure 5.8) and were driven by motors (such as at ‘50’) driving screw threads
‘53’ into attachments ‘5 1’.
Each pair of leaves was provided with a step at their inner faces so that when
and ‘60’ and figure
completely closed there is little leakage of radiation (see ‘45’
5.8).
Adjacent leaves on the same side of the collimator were also provided with a
step along the long edges for the same purpose.
The field can be verified by a TV arrangement, and is also seen projected onto
the skin surface in an analogous manner to conventional rectangular fields.
The collimator is shown in figure 5.9
-1.2.2. Ishigaki’s MLC

Multileaf collimators can be divided into two groups; those in which the leaves
cannot pass beyond the line joining the source to the isocentre, and those in which
they can. The latter are known as ‘over-running’ MLCs. Over-running MLCs allow
the field to be tailored to a target volume (or part of a target volume) through which
the axis of rotation does not pass. A collimator of this type has been described by
Ishigaki er a1 (1988, 1990a,b) from Nagoya University, Japan (figure 5.10). Table
5.1 summarizes its features.
This MLC was attached to a Mitsubishi ML- 15M 3 10MV linear accelerator and
used for both rotation therapy and multiple-fixed-port irradiation with irregularly
shaped fields and a maximum of 16 ports. The planning was CT guided and the
dose rate at each port could be varied. These authors did not state if this planning
was guided by optimization theory (see chapter 2). They constructed a phantom,

Figure 5.8. Detailed schematic of the method of controlling the positions of

opposing leaves in each pairfor the multileaf collimatordesigned by Brahme.
The leaves are shown at '30', '32', with faces '32', '45' moving on bearings
'36'-'38' and driven by screw threads '53' controlled by motors '50' acting
on connecting rods '51' . The rest of the diagram is supporting structure and
the spaces were filled with helium gas at atmospheric pressure. The leaves
are double-focused to the source (seefigure 5.7).(FromBrahme (1985/87).)
Figure 5.9. The multileaf collimator head and part of the light-freldpro-
jected on the patient using a 32 leaf-pair double-focusedcomputer-controlled
collimator of a medical microtron. The insert shows the patient contour and
the shape of the target volume in each CT slice. (From Brahme (1988).)
within which lay an irregular target volume. The planning problem demanded
the over-running leaf facility. By experimenting with different placements for the
rotation axis they found that the shape of each target area (defined as slices of the
volume) could be fitted well by the 90% isodose contour when a full 360" rotation
therapy was performed, and also by nine ports spaced at 40".
Ishigaki et a1 (1990a,b) gave the accuracy of leaf setting as 2 mm. They

Figure 5.10. View of the multileaf collimator developed by Ishigaki et a l .

There are I 1 pairs of leaves and each leaf can over-run by 5 cm. An example
of over-running is seen here. (From Ishigaki et a1 (1 988).)
used the overrunning collimators to delete a partial region within a field with
multiple-fixed-field irradiation, claiming the result was better than using a gravity-
assisted hanging-block method (see chapter 2). The computer-assistedfixed-field
irradiation, with irregular fields shaped using the overrunning collimators, can
produce more homogeneous target dose by changing the dose rate and deleting
partial regions within a field, thus shielding critical organs. There is no need for
the isocentre of rotation to be within the target volume.
5.2.3. Heidelberg’s MLCs

Boesecke et a1 (1 988) at the German Cancer Centre at Heidelberg have developed
an MLC with 28 pairs of tungsten leaves, each 4 mm wide, arranged to provide a
maximum field size of 16 cm at a distance of 1 m from the radiation source (figure
5.11). The collimator was constructed by the mechanical workshop of the Institute
of Nuclear Medicine at the German Cancer Research Centre (DKFZ).The aim
was to attach the MLC to a Siemens Mevatron 77 linear accelerator. By way of
contrast with some other MLCs, in which the MLC is an integral undetatchable
component, this one is accommodated in the accessory tray of the accelerator
and is therefore somewhat limited in size. The surface of each blade has no
step to prevent leakage radiation, but the leaves were double-focused to limit
penumbra. Two clinical versions of the collimator were made with the number
of leaves reduced to 27 pairs, each 3 mm wide (Schlegel 1992). One version of
the collimator (the manual version) uses a template and clamps to set the field
shape. In the motorized version there are six stepper motors: four control the
movements of four leaves, and the leaves to be moved are selected by translating
the leaf-moving assembly by two further stepper motors (Nahum and Rosenbloom
1989, Bortfeld 1992). Mechanical stresses, coupled with difficulties of coupling

Figure 5.11. The tungsten multileaf collimator with 28 leaves per side
developed at DKFZ Heidelberg, pictured in the mechanical workshop. This
multileaf collimator is not motor driven; the leaves are set manually. (From
Boesecke et a1 (1988).)
powerful stepper motors, have been problems to overcome with the motorized
version (Bortfeld 1992).
The manual Heidelbergcollimator was used in a fixed-fieldmode for treatments
with nine adapted field shapes spread over an angle of 330”. The field shapes
were determined so as to match the beam’s-eye-view of the target volume. This
volume comprised not only the tumour, but a margin to accommodate potental
tumour extensions and also positioning errors (figure 5.12). In turn, the beam’s-
eye-views were determined in a 3D treatment-planning package (also developed
at this centre) using the contours of the volumes of interest determined from x-ray
CT data. After planning, the isodose contours were displayed superposed on slices
of interest, which could be transaxial, sagittal or coronal. Additionally,the facility
to view three-dimensional isodose ‘ribbons’ superposed on three-dimensional
shaded-surface anatomy was possible (figure 5.13). Other clinical applications
were presented by Esik et a1 (1 99 1). The treatment-planningsoftware (known as
VOXELPLAN-Heidelberg)was described by Boesecke et a1 (1991).
The group at Heidelberg are currently developing an MLC with 1 mm wide
leaves for accurate conformation in the brain (Bortfeld 1992).
5.2.4. Kobayashi’s MLC

Kobayashi et af (1989) developed a multileaf collimator for a Toshiba LMR-
4C TOKU linear accelerator operating at 4 MV. The MLC comprised 11 pairs of
leaves, the inner nine being 25 mm wide and the outer two pairs being 75 mm wide.
The leaf-pairs were able to over-run the mid-line to the rotation centre by some 75
mm. In addition,on the source side of the MLC, primary collimation was provided,

Figure 5.12. The shape of the field generated for a multileafcollimator for
an irregularly shaped target volume. The projection of the target volume can
be seen as an array of black dots. The aperture of the collimator includes
a 'safety zone' to accomodate potential tumour extensions and possible
movement or positioning error. (From Boesecke et a1 ( I 988).)
Figure 5.13. A three-dimensional representation of a target volume and

spinal chord with an 80% isodose (delivered by multiple-static-MLC-shaped
fie1ds)presented as ribbons. In the originalpicture the surface and the ribbons
were shown in different colours. Provided the &ace does not protrude
through the ribbons, one can be certain that the dose to the target has reached
the value labelling the isodose, or is above that. (From Boesecke et a1 (1988).)
and on the patient side a wedge could be arranged if required. The unusual
feature of this development was the incorporation of a movable wedge-shaped
compensating filter (figure 5.14). Each pair of MLC leaves was accompanied by

Beam source
Figure 5.14. Showing the position of the MLC in relation to the additional
collimating elements for a Toshiba LMR4C TOKU 4 MV linear accelerator.
The movingfilter is on the patient side of the assembly. (From Kobayashi et a1
(1989).) (Reprinted with permission from Pergamon Press Ltd, Onford, U K . )
a corresponding pair of wedge filters, constructed from aluminium with a wedge

angle of tan 8= 0.1. As the MLC leaves varied in position, the location of the
corresponding wedge filters was adjusted so that the tip of each filter just touched
the beam edge defined by the ipsilateral leaf (figure 5.15). This geometrical
arrangement ensured that the filter thickness varied linearly with increasing width
of the field defined by each leaf pair. For example, when the width of the x-ray
beam at the rotation centre was 200 mm, the filter was 76" thick, but when the
width of the beam was 100 mm, the filter was only 38 mm thick. Since the MLC
leaves were able to traverse the mid-line to the isocentre, the movable wedges
followed suit.
Kobayashi et a1 (1989) made measurements with film dosimetry for a number
of irregularly shaped target distributions (the MLC leaves being adjusted to match
the geometry) in round and oval phantoms. These experiments mimicked the way
the leaves (and wedge filters) would be adjusted in clinical practice to accomodate
irregular target volumes determined from multiple CT images. It was observed that
the dose uniformity within the volume of the target was better when the wedge-
filters were in use than when the x-ray field was shaped only by use of the MLC
leaves alone. This was particularly the case when the calculated dose distributions
were compared with the measured dose distributions for those parts of the target
volume which were smaller than the widest part of the field and/or off-centre.
The dose profiles were also sharper, showing that using the wedge-shaped filters
protected the surrounding normal tissues more effectively from a high dose of
irradiation.
The rationale for use of a variable-thickness compensator may lie in the well
known observation that the penetration of a beam decreases as the field size

Modern developments 23 1
Figure 5.15. Showing the relationships between the leaf movement and the
wedge-filter movement for the collimation shown in figure 5.14. The tip of
each wedge just touches the ipsilateral field edge. Four direrent positions
of the collimator-wedge combination are shown. (From Kobayashi et a1
(1989).) (Reprinted with permission from Pergamon Press Ltd, Onford, UK.)
increases for a given megavoltage radiation (Green and Williams 1983). This
is partly due to increased scatter as the field size increases.
5.2.5. Mantel and PerryS MLC

Mantel et a1 (1984) and Perry et a1 (1984) have also described the construction
of an MLC with 12 leaf pairs of width 14 mm, attachable to a Toshiba LMR-16
linear accelerator. One somewhat unusual feature of their system was that the
control motors and circuitry for activating the leaf movements were external to
the collimator head. The system was used for dynamic rotation therapy.
5.2.6. Aoki's MLC

Aoki et a1 (1987) created an integrated radiotherapy treatment system in which
imaging for planning and radiotherapy treatment were performed in the same
room with the patient on a swivelled couch. A Mitsubishi multileaf collimator
under computer control was installed. This was used for dynamic conformation
therapy (which the authors called multileaf offset rotation-MOR) and fixed-
port conformation (multileaf offset multiport-MOMP), but they also allowed
the use of several axes of rotation when the treatment-planning problem called
for it. The dynamic and static multi-axis techniques were dubbed multileaf
offset multiaxis rotation-MOMAR and multileaf offset multiaxis multiport-
MOMAMP. The development is also discussed in chapter 7 (section 7.3)
5.2.7. Uchiyama's MLC

An MLC with leaves 2 cm wide at the axis of rotation was also included by

Uchiyama et a1 (1 986, 1990) in a system for conformal therapy based on the

beam's-eye-view of the tumour. The 11 leaves could cross the mid-line by 5 cm.
5.2.8. Bauer's MLC

An MLC with 13 leaves was constructed at the University of Freiburg and used
for multiport irradiation (10 fields arranged over 360" degrees) of oesophageal
carcinoma (Bauer 1992). The leaves diverge in the longitudinal but not in the
lateral field direction.
5.2.9. Biggs' MLC

Another home-made MLC has been constructed at the Massachussetts General
Hospital (Biggs 1991). It has 44 pairs of tungsten leaves, independently motorized
and capable of taking up position in 15 s. The leaves are focused as a sheaf
lengthwise, but not in the orthogonal direction. Each leaf width projects to 8 mm
at the isocentre; the penetration through the interleaf space is 8%. The leaves can
only move a cm or two across the mid-line. There is no verification system as yet.
The drive shafts are rigid and make the housing for the collimator quite bulky. The
largest field size is 35 by 33 cm. The aim was to use the MLC to cut down on the
number of custom-cast blocks and possibly, in time, achieve dynamic therapy.
Biggs et ai (1991) gave their reasons for deciding on a single- rather than
a double-focusing set of leaves (figure 5.16). To ensure the penumbra was
acceptably small when very large fields were set, the leaves were focused in the
plane orthogonal to the plane of the leaves (i.e. viewed end on, the tangents to
the long edges of the leaves converged to the focus.) However, focusing was
eliminated in the plane of the leaves. Biggs et a1 (1991) conducted a series of
experiments comparing the penumbra produced by a lead block with a straight
edge, a lead collimator with an edge focused to the source, a lead block having
a straight edge with a single step of 0.5 mm, and finally with a lead block
having a straight edge with a single step of 1 mm. Comparison was also made
with the penumbra from a tungsten block with a straight edge. The width of
several penumbra (95%-50% 90%-50% 80%-50%) was measured for different
field sizes, at different depths in water for a 10 MV x-ray beam on a Clinac 18
accelerator. No statistically significant differences in penumbra were found at
depth, whichever geometrical arrangements were compared. Some differences
were noted at small depths in water.
This enabled Biggs et a1 (1991) to conclude that there was no need for the
leaves to move on the arc of a circle, nor to have rounded ends, simplifying the
construction of their multileaf collimator. The same conclusion (for fields less
than 15 by 15 cm ) was reached by Dries (1991).

Brahme’s theory of orientation 233
Figure 5.16. A so-called ‘single-focused ’ multileaf collimator, The leaves

are focused to the source when viewed ‘end-on’ (left), but move in a plane
and are not focused in the orthogonal direction (right). This arrangement
should be contrasted with that shown in figures 5.7and 5.8. (From Biggs et
a1 (19911.)
Figure 5.17. Showing the fitting of a planar target area with a multileaf
collimator. The dotted area is the excess region treated by the ith leaf. The
best orientation of the leaves relative to the area is that which minimizes the
sum of such dotted areas of excess. (From Brahme (1988)J
5.3. BRAHME’S THEORY OF ORIENTATION
Brahme (1988) provided the arguments to answer the question of the optimal
angulation of the MLC leaves (at some particular static orientation relative to the
target volume). This led to the result that for a simple convex target volume the
direction of motion of the leaves should be in the direction in which the volume
presents its smallest cross section, i.e. the longest diameter of the target should be
covered with as many leaves as possible. The argument can be best understood
from figure 5.17.
Let the solid line represent the boundary of the target region and the leaves be
of width w. Let f ( x ) describe the boundary. For leaf i the overdosage is
oi = l:’(f ( x ) - f (xi)) h. (5.1)
If the boundary is monotonic and reasonably slowly varying, then it may be Taylor
expanded about xi with only the first three terms retained. This gives

Integrating, we obtain
Oi = w ( f ’ (xi)w/2 + f” (Xi)w 2 / 6 ) (5.3)
To obtain the total overdosage, integrate the expression
0= / lr
Oi = ( f ’ (x) w/2 + f ” (x) w 2 / 6 )dx (5.4)
from one side of the target region to the other to obtain
From this we see that:

1. the error is proportional to leaf width to first order, as one may expect,
2. the error is proportional to the variation in the target area in the direction
of the leaf movement,
3. the leaves should be aligned to minimize the opening of the collimator from
the fully closed position.
The problem reduces to finding the optimum way of arranging the leaves so as
to minimize the volume (represented by an area ‘seen’ in the beam’s-eye-view) of
normal tissue outside the target volume. This unwanted area arises because target
regions do not have convenient stepped outlines and cannot be precisely matched
by MLCs with leaves of finite width. Indeed, by considering a large number of
possible ways of fitting an elliptical area with an MLC of leaf width 1 cm, Brahme
(1988) showed that poor arrangements could lead to up to 20% more area being
treated than necessary. However, even this poorer arrangement was not as bad as
making use of a rectangular field which would treat some 35% extra area. The
reader could draw other geometrical target areas and go through the same type
of investigation of possibilities leading to different figures for the overdosage,
but the principle would be the same: using an MLC reduces the dose to normal
tissue outside the target area. When several MLC-defined fields are combined, the
advantages are even greater.
The problem becomes worse if the target area is not convex. Then (see figure
5.18) the range of directions which the leaf motion can take is restricted to less
than 360”. The figure shows that as the concavity becomes deeper the range of
angles becomes smaller. Indeed, if the opening to the cavity is smaller than its
interior, it is no longer possible to fit the target area however small the leaf width
is. Such cases as these could only be accommodated by dividing the field into sub-
fields, collimated one after the other. The problem becomes even more difficult if
there are multiple concavities. Using sub-fields may create matching difficulties
and is inherently less attractive.

Optimized beam profiles 235
Figure 5.18. The possible range of orientation (solid arrows) of the

direction of movement of collimator leaves for convex and convexlconcave
outlines. As the concavity increases, so the range of possible anglesfalls. The
dashed lines are the tangents at the inflection points of the target outline. The
region in the bottom right-hand figure cannot be treated by a single setting of
the leaves of a multileaf collimator because the opening is narrower than the
region to be treated within the concavity. Two adjacent fields could overcome
this, each collimated separately by the multileaf collimator. (From Brahme
(I988).)
5.4. OPTIMIZED BEAM PROFILES
Kallman et a f (1988) developed a method for delivering a dose by scanning an

elementary slit beam from an MLC. The density of application of such a slit for
the collimator is computed from a prescription of the (2D) target area, based on
an iterative method. Let us define the three-dimensional dose distribution from
an elementary slit beam by d (r). This is the most elementary dose distribution
which can be generated from a multileaf collimator-that corresponding to all
the leaves shut with just one leaf marginally open. Such a dose distribution may
be calculated from the basic point-spread functions for monoenergetic photon
interactions (Ahnesjo et a f 1987). The dose distributionmay then be approximated
by a continuum of collimated elementary slit beams, each tumed on for a different
time, which is called the density of application. Suppose the density of application
of such slit beams is CP (r),then the dose is given by
D (r) = 111 @ (r’)d (r - r ’ ) d3#.
Here we assume that the elementary distribution is independent of position and

then this is a convolution integral (in three dimensions). Because D and d are
known, the density CP can be computed by deconvolution. Kallman et a1 (1988)
perform this iteratively via:
Oo(r) = a D ( r ) (5.7)

,
‘, \
Figure 5.19. Showing ( a ) a target volume (cancer of the cervix) with

graded doses to the involved regional lymph nodes, ( b )the irradiation density
computed by the method of Kallman et a1 , (c) the resulting dose distribution
according to the irradiation density in (b)for %o irradiation, and ( d ) the
resultant dose distribution for a j a t field. (From Kallman et a1 (1988).)
@k+l (r) = c ( @ k (r) f a ( D (r) - @ k (r)@ d ( r ) ) ) (5-8)

where the subscript k refers to the kth iteration, C is a positivity constraint operator
and a controls the speed of convergence. This is formally identical to iterative
methods to remove blurring from medical images (see e.g. Webb 1988). It is also
formally similar to the iterative methods introduced by the same authors (Lind and
Kallman 1990) used to perform ‘inverse computed tomography’ (see chapter 2),
but note that in that case the kernel d has a completely different meaning, being
the point irradiation density from rotation therapy.
The opening density @ is a measure of the dose monitor units that should be
delivered, or alternatively the length of time the collimator should be opened
locally, assuming constant accelerator dose rate. The irradiation could also be
accomplished by a moving elementary slit beam with the dwelling time at each
location given by @. An example of the results of this method is in figure 5.19.
Similar iterative techniques were used by Lind and Brahme (1985, 1987) to

How the leaf positions may be determined 231
IA I A m o x
-1 1
4
3
‘f 2
Figure 5.20. Schematic diagram of the projection of an MLC field edge with
the direction of travel of the leaves at 45’ to the desired field edge. The
area blocked by the MLC is indicated by shading. The leaf width is ‘ w ’ .
Line 1 represents the location of the field edge for ‘exteriorfield blocking’,
line 2 represents the edge of the field for ‘interior field blocking’, line 3
represents the field edge for ‘leaf centre insertion’, and line 4 represents
the location of the field edge for ‘variable insertion’ blocking strategy. The
sinusoid represents the approximate location of the 50% isodose contour. The
strategies are discussed in section 5.5. (From Zhu et a1 (1992).)
determine the optimum scanning density of pencil beams from the desired fluence
profile.
The use of multiple geometrically shaped multileaf-collimated radiation fields
for optimizing 3D planning is discussed in chapter 2 (Mom11 et al 1991, Mohan
1990, Webb 1990, 1991,1992).
5.5. HOW THE LEAF POSITIONS MAY BE DETERMINED
Since each MLC has leaves of a finite width, the shape of any collimated field has
irregular staircase-like edges. Imagine the MLC leaves approaching a straight-line
field edge at 45“,as shown in figure 5.20. The question arises of how to place the
leaves in relation to the field edge, and Zhu et a1 (1992) have provided the data
to answer this question. If the leaf width is ‘w’,the saw-tooth irregularity will
translate to a sinusoidal variation in dose at depth with the maximum amplitude
of the variation being
Amax =W / m . (5.9)
In practice, measurements of the dose variation, with film dosimetry for cast
cerrobend blocks, double-focused to simulate an MLC, showed that amplitude A
of the 50% isodose contour (see figure 5.20) was some 1.2 mm smaller than Am,
for 6 MV radiation and was some 1.4 mm smaller than A,, for 18 MV radiation.
The difference between this amplitude and the value of Amax was independent
of the width ‘w’ of the leaves, a particularly convenient and useful result. The
sharp serrations of the leaves were thus ‘rounded out’ in the dose at depth (as also
discussed by Brahme).
Several options for placing the MLC leaves naturally arise:
1. The inner comers of the leaves are tangent to the field edge. No leaves project

into the field. Zhu et a1 (1992) call this ‘exterior field blocking’; the entire target
volume is thus exposed, but this may irradiate too much adjacent normal tissue.
2. The outer comers of the leaves are tangent to the field edge. Zhu et a1 (1992)
call this ‘interior field blocking’; no normal tissue is exposed to the target dose, but,
since the leaves project into the field, some target locations may be underdosed.
3. The edge of the field bisects the ends of the leaves. Zhu et a1 (1992) call this
‘leaf-centre insertion’; the 50% contour wanders in and out of the field.
4. The edge of the field is set so the leaves protrude into the field by A,, minus
1.2 mm (6 MV) or minus 1.4 mm (18 MV). Zhu et a1 (1992) call this ‘variable
insertion’ and this guarantees that the 50% isodose contour will undulate up to and
outside the field margin.
Zhu et a1 (1992) investigated these four possibilities, confirming the predictions
by film dosimetry. They also showed that convolution dosimetry techniques (see
chapter 2) would predict the dose distributions to within 3 mm accuracy and take
account of variable field shape, scatter and electron transport.
5.6. SYSTEMS COMMERCIALLY AVAILABLE IN 1992
The system described in section 5.2.1, due to Brahme, is available commercially

via Scanditronix. That described by Aoki et a1 (1987) is available from Mitsubishi.
Two other systems are commercially available, although only a small number are
in service worldwide in 1992. These are from Varian and Philips.
5.6.1. The Varian MLC

The Varian MLC (figure 5.21) has 26 leaves, each of width 0.5 cm projecting to
1 cm at a 1 m isocentre (Varian 1992). The leaves can retract to 20 cm either
side of the mid-line, giving a maximum field of 40 cm by 26 cm. The leaves
may over-run by the full 20 cm, but, when they do so, no two leaves may differ in
translated position by more than 16 cm because of the limitations of the secondary
collimation which allows this large over-run. All leaves move in a plane (rather
than on the arc of a circle) and have rounded ends to limit penumbra. No two leaves
can close to more than within 2 mm at isocentre. The collimator is detachable
from the accelerator, which retains the conventional block tray and wedge support
mechanisms below it. Experiments with the shape of the leaf ends were performed
in collaboration with the University of Pennsylvania and the final design (partially
rounded, partially flat) gave a penumbra of some 4 mm at 6 and 15 MV for various
leaf positions (Galvin 1990). The average transmission of the leaves themselves
was 4% of the open field (the leaves were 5 cm thick) and the transmission through
the gaps between the leaves added a further 1%. The leaf sides were double-
stepped to minimize this extra penetration. All leaves were focused to the source
in the direction orthogonal to the leaf movement. The collimator weighs 200 lb.

Systems commercially available in 1992 239
Figure 5.21. The Varian multileaf collimator. (From Varian product

literature.)
The leaves can be positioned to an accuracy better than 1 mm with a

repeatability of 0.1 mm and presently the speed of leaf movement is 1.5 cm s-'.
The leaf position is verified by measuring potentiometers, rather than positive
identification with a TV system. At present the leaves cannot be moved when
the gantry is rotating, so, when calculating the times to reset fields, these have
to be the sum of sequential gantry rotation and leaf movement times. At present
the interaction with the planning process is under development. The MLC is an
integral part of the accelerator head, but is controlled by an independent computer.
There is a facility to enter the required leaf positions by digitizer tablet. An
outline can be traced and then one of three options exercised: either the leaves
completely expose the field within the contour, or completely shield the region
outside the contour, or an option in-between (Kay 1992). The user interface is not
yet standardized and is specific to one site. In time, the MLC interface will have
access to the same data-base as the linac control. The wedge can be orthogonal to
the direction of movement of the leaves.
Boyer er a1 (1991) have reported measurements of penumbra and transmission,
and also the stair-step shaping of the isodoses at depth.
5.6.2. The Philips MLC

The Philips MLC has 40 leaves, each projecting to 1 cm at an isocentre of 1 m.
The blades are 7 cm thick, giving a mean transmission of some 1.8% at 6 MV
and 2% at 20 MV. The leaves may over-run by 125 mm and can be positioned to
within an accuracy of 0.25 mm in 10 s. The leaves have rounded ends and move
in a plane. There is a single step between adjacent leaves and the transmission
through the step is a further 2.8% at 6 MV and 3.1% at 20 MV (Williams 1990,
Williams and Jordan 1991). In addition, there are four conventional collimators;

two for x-collimation and two for y-collimation (the direction of movement of the
MLC leaves). The y-collimator has transmission 12.3% at 6 MV and 15% at 20
MV, and the corresponding figures for the x-collimator are 1.2% and 1.4%. When
an irregular field is collimated, the total leakage will therefore depend on which
combinations of collimators come into play, and can be deduced by producting
the appropriate fractional leakages. In addition, scatter from the primary radiation
in the open field will of course contribute into these areas. Williams and Jordan
(1991) found that the 80% isodose contour was about 4 mm inside the leaf tips
and the 50% contour went through the mid-points of the ends of the leaves.
The leaf positions are verified by a TV system. The leaves diffusely reflect light
onto the face of a CCD TV camera. The source of light is a halogen bulb. The apex
of the wedge is parallel to the direction of movement of the leaves.
The Manchester group work out the dosimetry for this collimator using
integrations over experimentally determined empirical scatter functions (Hounsell
and Wilkinson 1990). The Philips collimator presently comes with three methods
of setting the field shapes (Jordan and Williams 1991):-
1. regular rectangular fields are the simplest to set,

2. predefined shapes can be set via eight parameters,
3. irregular fields require the input of some 80 parameters, essentially two for
each leaf.
Williams and Kane (199 1) found that the third method, time consuming though
it was, was needed for many of the fields set at Manchester.
Kirby and Williams (1991) used the MLC in conjunction with a Philips SRI-100
MVI system to set up the fields for conservative breast irradiation. They made
use of the ability to set fields where one edge corresponded to the central beam
(so-called ‘half-fields’) so that match-line problems were minimized between the
tangential and the AP fields (Kirby and Kane 1991).
Hounsell et a1 (1992) have described how the treatment field is determined
from either digitized simulator radiographs or a digitized image-intensifier video
signal or from a digitally reconstructed radiograph produced from CT data. The
Manchester group have created a method of image display which allows the
locations of the MLC leaves to be overlaid on the digital image-display system.
They describe how the leaves may be positioned to satisfy one or other of the
altemative strategies discussed by Zhu et a1 (1992). An additional use of these
digital data is the means to compare with digital portal images for position
verification (Moore 1992).
The leaf positions are held in a ‘prescription file’, which is passed to the
computer of the treatment unit. Hounsell et a1 (1992) have found that, with a
typical treatment preparation time of only some 30 minutes, the use of the MLC
has led to a high through-put of patients with conformal fields.
The second Philips MLC to be installed in the UK has recently aITiVed at the
Royal Marsden Hospital, Sutton and a third is at London’s Royal Free Hospital.

References 24 1
5.7. MULTILEAF COLLIMATORS FOR PROTON THERAPY
In chapter 4 we discussed the use of high-energy protons for radiation therapy.

Proton fields may be shaped geometrically by sweeping elementary pencils
of protons with scanning magnets, or by expanding a small-area field with a
combination of scattering foils and beam-stopper. In the former case, a multileaf
collimator can be used as a back-up collimator, as in the irradiation facility at
Chiba, Japan (Tsunemoto et a1 1985, Kanai et a1 1991). In the latter case, the
multileaf collimator is the only beam-shaping device, as in the facility at Tsukuba,
Japan (Tsunemoto et a1 1985).
5.8. SUMMARY
We should be somewhat surprised that it has taken so long to engineer multileaf

collimators. It is self-evident that avoiding primary radiation at non-target-volume
sites is desirable and that custom-built blocking is tedious (Spelbring et a1 1987).
However, the engineering difficulties should not be underestimated and those
centres who have chosen to build their own collimators have reported that it has
not been an easy task. The main engineering problems are machining tungsten
accurately (including the steps at the interfaces), arranging the motorization of the
leaf movements and verifying the leaf positions. The ‘interface’ to the planning
computer and to the treatment machine itself also has to be engineered. At the
time of writing, MLCs are still not widely available, so it is too early to make
authoritative statements concerning their utility. With an eye of faith, one expects
to see them become the standard treatment accessory to the machine head.
REFERENCES
Ahnesja A, Andreo P and Brahme A 1987 Calculation and application of point

spread functions for treatment planning with high energy photon beams Acta
Oncol. 26 49-56
Aoki Y,Akanuma A, Karasawa K, Sakata K, Nakagawa K, Muta N, Onogi Y
and Iio M 1987 An integrated radiotherapy treatment system and its clinical
application Rad. Med. 5 131-141
Bauer J, Hodapp N and Frommhold H 1992 A comparison of dose distributions
with and without multileaf-collimator for the treatment of oesophageal
carcinoma Proc. ART91 (Munich) (abstract book p 49) Advanced Radiation
Therapy: Tumour Response Monitoring and Treatment Planning ed A Breit
(Berlin: Springer) pp 649-653
Bewley D K 1989 The physics and radiobiology of fast neutron beams (Bristol:
Adam Hilger) pp 69-7 1
Biggs P 1991 private communication (AAPM travelling fellow visit, May 21st
1991)

Biggs P, Capalucci J and Russell M 1991 Comparison of the penumbra between

focused and nondivergent blocks-implications for multileaf collimators Med.
Phys. 18 (4) 753-758
Boesecke R, Becker G, Alandt K, Pastyr 0, Doll J, Schlegel W and Lorenz W
J 1991 Modification of a three-dimensional treatment planning system for the
use of multi-leaf collimators in conformation radiotherapy Radiother. Oncol.
21 261-268
Boesecke R, Doll J, Bauer B, Schlegel W, Pastyr 0 and Lorenz W J 1988
Treatment planning for conformation therapy using a multileaf collimator
Strahlentherapie und Onkologie 164 151-154
Bortfeld T 1992 private communication
Boyer A L, Nyerick C E and Ochran T G 1991 Field margin characteristics of a
multileaf collimator Med. Phys. 18 606
Brahme A 1984 Multileaf collimation of neutron, photon and electron beams Proc.
3rd Ann. ESTRO meeting (Jerusalem, 1984) p 2 18
-1985/87 Multileaf collimator US Patent no 4672212
- 1987 Design principles and clinical possibilities with a new generation of
radiation therapy equipment Acta Oncologica 26 403-412
-1988 Optimal setting of multileaf collimators in stationary beam radiation
therapy Strahlentherapie und Onkologie 164 343-350
Chu J C H and Bloch P 1986 Static multileaf collimator for fast neutron therapy
Med. Phys. 13 578
-1987 Static multileaf collimator for fast neutron therapy Med. Phys. 14 289-
290
Dries W J F 1991 Comparison between the penumbra of focused and straight
standard shielding blocks Proc. 1st biennial ESTRO meeting on physics in
Eenmaa J, Kalet I and Wootton P 1985 Dosimetric characteristics of the
University-of-Washington clinical neutron therapy system (CNTS) multileaf
variable collimator Med. Phys. 12 545
gsik 0, Burkelbach J, Boesecke R, Schlegel W, Nmeth G and Lorenz W J 1991
3-dimensional photon radiotherapy planning for laryngeal and hypopharyngeal
cancers: 2 Conformation treatment planning using a multileaf collimator
Radiother. Oncol. 20 238-244
Galvin J 1990 private communication
Green D T and McColm C S 1952-54 Beam therapy collimating unit U S Patent
no 2675486
Green D and Williams P C 1983 Brit. J. Radiol. (Suppl 17) 61-86
Gscheidlen W 1959 Device for collimation of a ray beam US Patent no 2904692
Hounsell A R and Wilkinson J M 1990 Data for dose computation in treatments
with multileaf collimators The use of computers in radiation therapy:
Proceedings of the 10th Conference ed S Hukku and P S Iyer (Lucknow) pp
40-43
Hounsell A R, Sharrock P J, Moore C J, Shaw A J, Wilkinson J M and Williams

References 243
P C 1992 Computer-assisted generation of multi-leaf collimator settings for

conformation therapy Brit. J. Radiol. 65 321-326
Ishigaki T, Itoh Y, Horikawa Y, Kobayashi H, Obata Y and Sakuma S
1990a Computer assisted conformation radiotherapy The use of computers in
Iyer (Lucknow) pp 76-79
-1990b Computer assisted conformation radiotherapy Med. Phys. Bulletin of
the Association of Medical Physicists of India 15 (3,4) 185-189
Ishigaki T, Sakuma S and Watanabe M 1988 Computer-assisted rotation and
multiple stationary irradiation technique Europ. J. Radiol. 8 76-8 1
Jordan T J and Williams P C 1991 Commissioning and use of multi-leaf collimator
system Proc. 1st biennial ESTRO meeting on physics in clinical radiotherapy
(Budapest,1991) p 3
Kdlman P, Lind B, Eklof A and Brahme A 1988 Shaping of arbitrary dose
distributions by dynamic multileaf collimation. Phys. Med. B i d . 33 1291-1300
Kanai T, Kawachi K and Hiraoka T 1991 An irradiation facility and a beam
simulation program for proton radiation therapy Nucl. Znstrum. Methods in
Phys. Res. A 302 158-164
Kay A 1992 Varian (Crawley) Open Day Lecture. February 21st 1992
Kirby M C and Kane B 1991 Megavoltage imaging-its role in breast treatment
verification Rad. Magazine (Sept 1991) 17 (196) 22-23
Kirby M C and Williams P C 1991 Initial experience of megavoltage imaging
for breast treatment verification Proc. Radiology and Oncology 91 (Brighton,
1991) (London: BIR) p 82
Kobayashi H, Sakuma S , Kaii 0 and Yogo H 1989 Computer-assisted
conformation radiotherapy with a variable thickness multileaf filter Znt. J.Rad
Oncol. B i d . Phys. 16 1631-1635
Leavitt D D, Gibbs FA, Heilbrun M P, Moeller J H and Takach G A 1991 Dynamic
field shaping to optimise stereotactic radiosurgery Znt. J . Rad. Oncol. Biol.
Phys. 21 1247-1255
Lind B and Brahme A 1985 Generation of desired dose distributions with scanned
elementary beams by deconvolution methods Med. Bio. Eng. and Comp. 23
(Suppl2) 953
- 1987 Optimisation of radiation therapy dose distributions using scanned
electron and photon beams and multileaf collimators The use of computers in
radiation therapy ed Bruinvis et a1 (Proc. 9th ICCRT) pp 235-239
Lind B K and Ktillman P 1990 Experimental verification of an algorithm for
inverse radiation therapy planning Radiother. Oncol. 17 359-368
Maas W and Alexandrescu M 1982 Rontgenuntersuchungsgerat mit einer
femsteuerbaren Primbstrahlenblende German patent app DE 3030332 AI
Mantel J, Lefkofsky M A, Higden D and Perry H 1984 Use of a multileaf
collimator for dynamic rotation therapy Proc. 3rd Ann. ESTRO meeting
(Jerusalem, 1984) p 219
McGunnigle H A 1926-29 Apparatus for altering the shape and size of a projected

beam of light US Patent no 1709626

Mohan R 1990 Clinically relevant optimisation of 3D conformal treatments The
S Hukku and P S Iyer (Lucknow) pp 3 6 3 9
Moore C J 1992 Correlation of megavoltage and simulator images Proc. BZR Conf.
on Megavoltage Portal Imaging (London, 1992) (London: BIR)
Mom11 S M, Lane R G and Rosen I 1991 Constrained simulated annealing for
optimised radiation therapy treatment planning Comp. Meth. Progr. Biomed.
33 135-144
Nahum A and Rosenbloom M E 1989 private communication (notes on a visit to
Heidelberg, July 1989)
Perry H, Mantel J, Vieau D M, Higden D L and Lefkofsky M M 1984 Dynamic
treatment: automation of a multileaf collimator Proc. 8th Znt. Con& on the Use
of Computers in Radiation Therapy (Toronto, 1984) (IEEE Computer Society,
Toronto) pp 479481
Robinsohn I 1906 Apparat zur radiologischen Belichtung lebender und lebloser
Objekte German patent no 192300
Schlegel W 1992 private communication to S Webb (visit to Heidelberg, October
1992)
Spelbring D R, Chen G T Y,Pelizzari CA, Awan A and Sutton H 1987 A computer
simulated evaluation of multivane collimator beam delivery systems The use of
computers in radiation therapy ed Bruinvis et a1 (Proc. 9th ICCRT) pp 153-156
Tsunemoto H, Morita S, Ishikawa T, Furukawa S, Kawachi K, Kanai T, Ohara
H, Kitagawa T and Inada T 1985 Proton therapy in Japan Rad. Res. 8 (Suppl)
S235-S243
Uchiyama Y, Kimura C, Ueda T, Morita K and Watanabe M 1986 Systematization
of radiotherapy units by computer system Rad. Syst. Res. Suppl. (Suppl. 3) pp
2 17-220
Uchiyama Y, Morita K, Kimura C, Mizutani T and Fuwa N 1990 Integrated system
of the computer-controlled conformation radiotherapy The use of computers in
Iyer (Lucknow) pp 52-55
Varian 1992 Varian’s multileaf collimator Varian Associates product literature
Wambersie A 1990 Present and future work at the radiotherapy and radiobiology
departments of the UCL-St Luc University Hospital, Brussels ICRU News 2
(90) (Dec 1990) 16-18
Webb S 1988 The mathematics of image formation and image processing The
physics of medical imaging ed S Webb (Bristol: Adam Hilger) pp 534-566
-1990 A new dose optimising technique using simulated annealing Proc. 9th
Ann. Meeting of ESTRO (Montecatini,1990) p 263
-1991 Optimisation by simulated annealing of three-dimensional conformal
treatment planning for radiation fields defined by a multileaf collimator Phys.
Med. B i d . 36 1201-1226
-1992 Optimisation by simulated annealing of three-dimensional conformal
treatment planning for radiation fields defined by a multileaf collimator. 2:

References 245
Inclusion of two-dimensional modulation of the x-ray intensity Phys. Med.

Biol. 37 1689-1704
Williams P C 1990 private communication
-1992 Shaping the treatment volume Proc. 50th Ann. Congress of the British
Institute of Radiology (Birmingham, 1992) (London: BIR) p 15
Williams P C and Jordan T J 1991 Performance characteristics of the Philips
multileaf collimator Proc. Radiofogy and Oncology 91 (Brighton, 1991)
(London: BIR) p 83
Williams P C and Kane B 1991 Use of predefined shapes for producing irregular
fields with a multileaf collimator Proc. Radiofogyand Oncology 91 (Brighton,
1991) (London: BIR) p 82
Zhu Y, Boyer A L and Desobry G E 1992 Dose distributions of x-ray fields shaped
with multileaf collimators Phys. Med. Biol. 37 163-173

CHAPTER 6
MEGAVOLTAGE PORTAL IMAGING
6.1. THE NEED FOR HIGH-QUALITY PORTAL IMAGING
It must be continually emphasized that precision in radiotherapy depends on the

accuracy of a sequence of stages in the planning and delivery phases; since a
chain is only as strong as its weakest link it is important that no aspect of the
process receives less attention than the others. Hendee (1978) prophetically wrote:
‘With improved dose-perturbation correction techniques, will not the accuracy
of dose distributions be determined primarily by how well the patient geometry
used to collect CT data simulates the patient geometry employed for radiation
therapy?’ Until comparatively recently, portal imaging for ensuring the accuracy
of positioning ut the time of treatment was not a very active focus of attention.
Despite many obvious deficiencies, film was the widely favoured detector of
megavoltage photons. Although improving portal imaging does not solely benefit
conformal radiotherapy, it has received greater attention as conformal therapy
becomes a more realistic prospect. This is possibly because of the statement
that conformal therapy needs to be more accurate, in the sense that the tighter
the dose distribution to the target, the more serious patient mispositioning in the
treatment fields will become (Swindell and Gildersleve 1991). This is discussed
philosophically by Smith (1990).
The main limitations of film are:
0 It has low contrast for megavoltage radiation. Scattered radiation contributes to
this low contrast.
0 The images are not taken in ‘real-time’. Development times are relatively long.
This prohibits using the images to adjust the position of the patient prior to
treatment.
0 The data are analogue. This precludes more than qualitative comparisons with
simulator and DRR images.

0 The cost of the film is largely irrecoverable.
A number of altemative detectors have been developed which each overcome

some or all of these limitations. For example, digital images not only open up
possibilities for quantitative comparison (e.g. registration) with simulator images
or digital projections from CT data, but can also be processed (e.g. by windowing,
246
Fluoroscopic detectors 247
Figure 6.1. An electronic portal image (EPI) made by a digital portal imager
(E) on a linear accelerator (left) may be compared with digitized film (DF)
taken on a radiotherapy simulator (centre) or with a digitally reconstructed
radiograph (DRR) created from x-ray CT (C) data (right). Each image may
have a different magnifrcation and pixel size which must be corrected before
quantitative comparisons are made. These comparisons assess how well the
patient positioning is maintained through imaging, simulation and treatment.
histogram equalization etc) to overcome poor contrast (Smith 1987) (figure 6.1).
Imaging can be performed in real-time, allowing intervention and intelligent use
of the data. The cost is transferred to the development phase and/or the capital
outlay, each image being essentially ‘free’. An argument that film is at least hard-
copy can now be easily overcome as computer-driven hard-copy devices become
readily available.
A variety of technologies have emerged and are naturally somewhat competi-
tive. There are pioneering ‘home-made’ systems, as well as commercially avail-
able products. There is plenty of debate over which is the ‘best’ system (figure
6.2). The main developments have been: fluoroscopic area detectors, scanning-
photodiode and scintillator-plus-photodiodearrays, matrix ionization chambers,
photostimulable phosphors, xeroradiography and silicon area detectors (Boyer et
af 1992). The main features of each will now be described.
6.2. FLUOROSCOPIC DETECTORS
Fluoroscopic detectors were the earliest developments, being the natural analogue
of the fluoroscopic imager so common on radiotherapy simulators. Benner et a1
(1962) claimed to be the first to image very high-energy (30 MV) radiation with
a television fluoroscopic technique. The energetic photons were generated by a
Betatron and imaged on to a 1.5 mm thick lead sheet bonded to a layer of large
ZnCdS crystals. The lead sheet converted photons into electrons, which generated
light in the fluorescent material with a very small, but not negligible, efficiency.
The low-luminance photon emission was sensed by an experimental prototype
orthocon television camera.
The very characteristics which make high-energy photons desirable for
radiotherapy, namely their lack of differential absorption between bone and soft
tissue and their small attenuation, minimizing the effects of tissue inhomogeneities

248 Megavoltage portal imaging
Figure 6.2. The two main classes of electronic portal imager. Left-a
one-dimensional array of radiation detectors (D)scans across the radiation
field ( F ) and builds up the image on a monitor ( M )from a series of (1 x N j
vectors of pixels. The radiation ( W ) not falling on the detector is wasted,
although each part of the radiation field is of course sampled in turn as the
detector scans (arrow). A detector of this type must have a high quantum
efficiency. The detector is shown schematically and in practice has many
elements and may have two rows of elements, the rows offset by halfa detector
spacing to improve spatial resolution. Right-a two-dimensional detector
(0)imaging the whole field (F).The image is displayedas an (N x N ) matrix
of pixels on a monitor ( M I . Such area detectors generally have a much poorer
quantum efficiency. Both detectors are shown at a large distance from the
isocentre, which reduces the contribution from scattered radiation.
and improving depth dose, are the enemies of megavoltage imaging. To overcome
this Benner et a f (1962) had the patient swallow a plastic tube containing lead, gold
or tungsten to act as a visible landmark with some contrast! Gold surgical clips
served the same purpose for other anatomical sites. Modem portal imaging shows
artificial-limb implants effectively for the same reasons.
To put these developments in perspective, it should be recalled that it was
common practice to set up a patient using images with diagnostic x-rays, followed
by wheeling the patient (supposedly without moving on the couch) to the therapy
set in the same room (figure 6.3).
Baily et a1 (1980) (at the University of Califomia) and Leong (1986) (at
the Massachusetts General Hospital, Boston) describe systems with similar
characteristic features. In Baily’s system, megavoltage x-rays are incident on an
area detector comprising a &” stainless-steel plate cemented to an E-2 fluorescent
screen (Dupont). Baily’s detector was 17” x 17”. The light from the scintillator

Figure 6.3. Showing an old practice of having a diagnostic x-ray set for
arranging the patient position in the same room as the treatment machine.
The patient was wheeled unmoving between the two machines. (FromSiemens
Reineger Werke News, July 1957.)
was bent through 90" by a front-surfaced mirror on to a video system via a 13

mm focal-length lens, a silicon-intensified target (SIT) Vidicon tube and a video
camera. The 90" bend is necessary to avoid irradiating the camera. The limiting
resolution (for 6 MV radiation) in Baily's system was 0.8 lp mm-', measured at
the surface of the fluorescent screen. The system was used to image 6 MV and
6oCoradiation. Leong passed the video output into a real-time digital frame store
capable of storing 512 x 512 images, either 8 or 16 bits deep at 30 frames per
second (see figure 6.4). This, combined with a pipelined image processor, enabled
numerical operations on this data at the same rate as frames were acquired. For
example, time averaging (integration) for noise reduction, discrete convolutions,
decalibration, grey-scale manipulation etc, were all possible.
The system, in addition to creating digital images, also acted as a differential
dosemeter (single frames digitized to 8 bits) or integral dosemeter (summed
frames digitized to 16 bits). In the latter mode a theoretical dynamic range of
65 000 was possible. Leong compared the output of the detector as a differential
dosemeter with that of a Farmer ionization chamber showing linearity over
.
doserates from 100to 500 cGy min-' He also compared the output of the detector
as an integrating dosemeter with the monitor ionization chamber of the treatment
machine, also showing linearity with dose. Both these comparisons were made
with only 2500 detector pixels, to avoid the need to consider detector spatial
non-uniformity. Furthermore, the system compared favourably with film when
measuring wedged fields (figure 6.5). There were no saturation effects. The

Imoging section Imoge processing section

I 1
A i e x-roy
o m
E-2 Image
fluorescent-- processor
screen
Mirror
- TRAPlX
VAX111780
b
3 0 1 1 , , , , 1 I , , I , , , , , ,
-8 -6 -1 -2 0 2 L 6 8
cm
Figure 6.5. The curves show the performance of Leong’s portal imaging
system acting as a dosemeter compared with film (KodakXV2)for two wedged
fields. The solid lines show the jilm response and the single points are from
the digital fluoroscopy system. (From Leong (19861.)
system could thus be used with confidence to map the exit dose from a treatment
portal. In integrating mode it could be used to become a total-dose monitor and act
as a ‘verify’ system. For example, it could be coupled to switch off the radiation
beam if too many monitor units had been set, and could detect incorrect wedges.
If there were a need for extensive image processing, e.g. frame averaging, the
images themselves took some 2-4 s to be presented on a TV monitor.
Leong provided an early indication that the human observer was possibly the
new ‘weak link in the chain’. Could such an observer detect errors rapidly? Could
these be rapidly corrected? These questions are still debated.
Shalev et a1 (1989), in Winnipeg, have also developed on-line portal imaging

Fluoroscopic detectors 25 1
using a metal-plus-phosphor detector viewed via a 45” mirror by a TV camera.

They used an SIT video camera and obtained images by averaging 128 or 256
frames (4-8 s acquisition time) and post-processing the data (Leszczynski et a1
1988, 1990). A major consideration with video-based MVI systems is the need to
optimise the transfer characteristics of each stage of the system; for example there
are very large light losses in the optical coupling between scintillator and TV. If
the magnification is m, the refractive index of the scintillator is n , the F-number
of the lens is F and its transmission is T , then the optical transmission factor is
2
which is very small for practical values of the variables (Swindell 1991). Shalev
has a factor ‘4’not ‘16n2’in the above equation. Others (e.g. Munro et a1 199Oa)
have ‘8’. The correct factor is ‘16n2’,because the ‘object’ has a refractive index
and is self-luminant, rather than reflecting incident light in a Lambertian manner
(for which the ‘usual’ formula with the ‘4’ was developed. Under these different
circumstances the lens transmission factor is simply the fractional geometrical
solid angle collecting the light producted with the transmission of the glass, and
the ‘4’ is appropriate.) Swindell (1991) has clearly distinguished between these
conditions.
The fluoroscopic imaging system can be considered as a ‘cascade or chain of
events’. The output signal-to-noise ratio SNROutis related to the input signal-to-
noise ratio SN Ri, via (Barrett and Swindell 1981) the equation:
SNR~P~
SNRLt = (1 + (l/P3P4g2))
where
0 PI is the probability of an x-ray interacting with the metal converter plate
and producing a high-energy electron which enters the phosphor (the quantum
efficiency of the metal plate),
0 g2 is the number of photons emitted from the screen per absorbed electron,
0 P 3 is the probability that a photon generated within the screen will pass through
the lens and be incident on the photocathode of the TV camera,
0 P 4 is the probability that a light photon will create a photoelectron in the target
of the TV camera.
This analysis assumes the conversion of high-energy electrons to light photons
is a Poisson process and all other stages in the cascade are binary selections. The
product P3 P4g2 is all important, being the number of light photons detected by the
TV camera for every x-ray photon interacting in the metal plate. It is necessary for
this figure to be as high as possible so that the noise is not amplified by the imaging
system. g2 is determined by the type of screen and its thickness. Unfortunately
P3 is usually very small.

Set against this large loss factor due to P3, an area detector has some advantages
over a scanning linear array (see section 6.3). The area detector has a very poor
quantum efficiency ( P I is only about 1% (Shalev 1991, Shalev et a1 1988)) but of
course captures data over the whole field of view simultaneously. The scanning
linear array, to compensate, needs a very high quantum efficiency of the order 50%
or more, and, to this end, thick high-density crystals are used. The area detector
can make real-time movies (Shalev et a1 1988); the scanning linear array can only
make movies with a time periodicity of the scan.
Munro et a1 (1990a,b) (in Toronto) have constructed a digital fluoroscopic
megavoltage imaging system for radiotherapy localization, paying careful
attention to optimizing the features of the design which lead to noise propagation.
Munro et a1 (1990b) estimate P3 = 2.8 x for a F=0.95 lens and a
demagnification of 20 and lens transmission coefficient of 0.9. Munro et a1
(1990b) have an ‘8’ rather than the correct ‘16n2’ in the denominator of the
expression (see equation (6.1)) for P3 and the correct value is therefore at least
a factor 2 smaller than stated. To counter the small P3, Munro et a1 (1990b) use a
thick phosphor screen (400 mg cm-2) which gives a light output measured to be
g2 = 2.5 x lo4 light photons per high-energy electron. With P4 estimated to be
0.75, the product P3 P4g2 turns out to be about five. It might be a factor of five
smaller, however, due to other uncertainties. Munro et a1 (1990b) considered six
different thicknesses of phosphor, showing the light output was roughly linear with
thickness (there is a small non-linearity due to optical attenuation in the phosphor)
and the use of thick gadolinium oxysulphide phosphor bonded to a 1 mm thick
copper plate was one of two important improvements they made to this type of
system.
The poor quantum efficiency of the area detector and the potential degradation
of SNR,,, by a small P3P4g2 product has to be compensated by video frame
averaging. In the Winnipeg system (Leszczynski et a1 1988) static images
have signal-to-noise ratio improved by a factor of 16 by averaging 256 frames.
However, for making movies the acquisition time has to be shortened to some s
so averaging over only 16 frames is possible, which improves the signal-to-noise
ratio by only a factor of four. Additional non-uniformity corrections and contrast
enhancement by adaptive histogram equalization were also needed. This system
is being evaluated clinically (Shalev et a1 1991).
The second major improvement made by Munro et a1 (1990b) to overcome the
effects of noise was to arrange for some of the integration to be performed on the
TV target rather than in the frame grabber. Their electronics allowed two modes
of operation:
1. Frames grabbed at 30 frames per s continuously for 256 frames (total
imaging time 8.5s).
2. Charge integrated for up to 2 s on the TV target with sequential frames
integrated in the framestore of the framegrabber. As the latter could hold 256
frames, this second mode was capable of imaging over a period as long as 8.5
minutes, i.e. throughout an irradiation.

In the second mode it can be shown that the noise contributed by the TV camera
itself is minimized (since it only contributes on data readout and the number
of readouts is & of that in the first mode.) The shot noise (included in P4) is
proportional to the square root of the signal: thermal and amplifier noise can be
ignored by comparison. A potential disadvantage of mode 2 is that the dark current
increases proportionally to the length of time the signal is accumulated, but Munro
et a1 (1990b) show this is always less than 5 % of the real signal and can essentially
be ignored.
Experimentally, they showed that the use of these two improvements (optimized
phosphor screen and integration on the TV target) lead to much better imaging in
practice, with contrast as low as 0.7% visible with a 7 cGy irradiation. Further
improvements might arise if PI can be optimized. There are no reliable estimates
of the electron production from high-energy x-rays interacting in thin metal plates,
although studies are in progress (Wowk et a1 1991).
Visser et a1 (1990) have developed a fluoroscopic system that is marketed by
Philips. The prototype has been in use at the Daniel den Hoed Cancer Centre
since July 1988. A demountable 30 x 40 cm2 rare-earth fluorescent screen (411
mg cm-2) is viewed via a 45” mirror by a CCD camera permanently installed on
the linac. The field at the isocentre is 19 x 25 cm2 maximum. Integration is
performed in the CCD camera for between 0.1 and 1 s with further integration in
the frame store. The system is calibrated daily with an open-field measurement.
Images may be filtered for edge enhancement and histogram equalization and also
grey-scale windowed. With a 7 s exposure and a 25 cm absorber present, a dose of
42 cGy allows a signal-to-noise ratio of 105. The spatial resolution was measured
as 1.5 mm FWHM (compared with 0.87 mm FWHM for film). Measurements with
a Lutz and Bjarngard (1985) test phantom gave 0.74% contrast visible.
Fluoroscopic portal-imaging systems have been developed by several com-
mercial manufacturers. The system of Shalev and colleagues evolved into the
‘Beamview Plus’ product from Siemens, which creates a 26 cm by 33 cm image
at the isocentre with some 0.5-3 cGy dose depending on the site. A feature of
the system is that the imager can be telescoped into the gantry for ‘parking’. The
system of Munro and colleagues evolved into the ‘Theraview’ product from S and
S Inficon. This also has a retractable bellows for parking the imager. Images with
a spatial resolution of 1 lp mm-’ and contrast resolution of 0.5% are formed with
1-2 cGy dose. The imager has a 40 cm by 40 cm metal plate and the image size
at isocentre obviously depends on the geometry; the detector can move vertically
as well as be retracted. The Philips system is known as ‘SRI-100 MVI’ and has a
rigid frame without the option to retract into the gantry.
One of the major disadvantages of all the systems discussed so far is their
physical bulk. A group at the Mallinckrodt Institute of Radiology have developed
an optical coupling which does away with the mirror and is inherently more
compact. Wong et a1 (1990a, 1991) linked a 40 cm by 40 cm input screen to a
3 cm by 3 cm output plane via bundles of fibre-optic image reducers; 256* plastic
fibres were used. At one end these were coupled to a fluorescent screen with 300

Photon Converter esccnce Screen
Figure 6.6. The prototype optical coupling in the fibre-optic imager. The
jluorescent screen is coupled to the front face of a digital camera by bundles
of fibres. This dispenses with the need for a 45" mirror and reduces the bulk
of the detector. (From Wong et a1 (1990).) (Reprinted with permission from
Pergamon Press Ltd, Odord, UK.)
mg cm-2 of gadolinium oxysulphide beneath a 3 mm copper sheet. At the other

end they coupled to the input of a TV camera with a digital 5122 pixel output
(figure 6.6). The fibres were in bundles of 162, each bundle optically isolated
from its neighbours to reduce inter-bundle cross-talk. The bundles tumed through
a right angle within 10 cm of the input screen, leading to a shallow device. The
efficiency of the fibre coupling system is much the same as for an efficient lens
system (Boyer et a1 1992). This is because nearly all the optical photons are tumed
around in the tapering light-pipe due to multiple reflection (Swindell 1992b).
The raw images suffered considerable spatial distortion and non-uniformity
of more than 60%, both effects needing software correction. The usual digital
windowing and histogramming facilities were available. Typically 8 to 128 frames
of video data were averaged to generate each image. Despite these corrections,
the prototype system was judged poorer than optimal with a spatial resolution
of 0.3 line pairs per mm and a density resolution of 1% for an object 1.5 cm in
diameter. The non-uniformity of the light-pipes could have been reduced by pre-
selecting only those fibres with common optical properties. Nevertheless, useful
phantom and clinical images were obtained. Indeed Graham et a1 (1991) and Jones
and Boyer (1 99 1) have reported using the system (marketed as 'Fibervision') to
assess patient movement during therapy. The latter developed a Fourier-based
correlation technique to compare digital portal images with simulator images.
6.3. SCANNING LINEAR ARRAY OF DIODES AND CRYSTAL

DETECTORS
The scanning-diode array for megavoltage radiation imaging originated with

Taborsky et a1 (1982) and was further developed by Lam et a1 (1986). Taborsky

Scanning linear array of diodes and crystal detectors 255
et a1 (1982) performed a design study to show the feasibility of using a scanning

linear array of photodiodes as a portal detector. In the design study the diode array
was fixed in space and the object to be imaged was translated through the field.
The detectors were silicon (Z=14) photodiodes of radius 1 mm embedded in 2
mm thick lead. About 11% of the (4 MeV) x-rays interacted with the leaddiode
detector, the rest passing straight through undetected. A Monte Carlo calculation
of the self absorption by the lead of the secondary Compton electrons showed that
some 30% only were detected by the photodiode. Thus the quantum efficiency of
the detector was some 3%. They estimated that each accelerator pulse delivered
1.77 x lo7 photons cm-2. Thus for each pulse some 1.77 x 1071s(0.1)20.03=
1.83 x lo4 primary photons were detected, from which the fractional noise from
the photon statistics alone was some 0.7%. They estimated electronic noise as
some 0.5% and so the total signal-to-noise ratio was 116. More complicated
calculations, taking into account the variable output per accelerator pulse, showed
the signal-to-noise ratio was 119 for an open field. The effect of this variability
was calibrated out by mounting a detector of five diodes close to the accelerator
collimator. Taborsky et a1 (1982) built a small test prototype with 62 diode pairs
(mounted one above the other and connected in parallel to double the signal per
2 mm element of detector) and used it to make small images with a 62 x 67 matrix
of sample points, displayed with just eight grey levels. They also noted that other
room-temperature detectors with higher conversion efficiencies could have been
used, such as cadmium telluride (Z=50) and mercuric iodide (Z=62).
Lam et a1 (1986) (at Johns Hopkins Medical Institute, Baltimore) constructed
a linear array of 255 silicon diodes, with the centre-to-centre separation between
diodes of 2 mm. A layer of lead 1.1 mm thick was placed above each diode for
electronic build-up. The line of diodes was scanned under computer control, the
step size between scan lines being 2 mm. The maximum number of scan lines
of 224 gave a maximum field size of 44.8 cm (in the scan direction) by 51 cm
orthogonal to this. The cathodes of all diodes were grounded and the anode of each
diode was connected to a charge amplifier and thence to a sample-and-hold unit.
The output voltage of each assembly was proportional to the charge generated
at the diode by the impinging x-rays. The use of sample-and-hold ensured the
signals detected at all diodes were sampled simultaneously and held at the peak
voltage. The computer which controlled the stepper motor governing scanning
also accepted the signals which were multiplexed and passed to a single analogue-
to-digital converter, working at 12-bit resolution (figures 6.7 and 6.8).
After scanning, the signal level of each diode was corrected for variable
detection efficiency and background noise using prestored data obtained by
scanning an open field. A further single photodiode placed near the collimators
of the accelerator was used to calibrate the image for intensity fluctuations in the
x-ray beam. The final image was video-reversed (so the image resembled a film
image with dark corresponding to minimum photon absorption) and displayed as
a 256* or 5122 image with 8 bits (256 grey levels). The detector was 50 cm below
the isocentre (at 1 m) giving a spatial resolution of 1.3 mm and measurements

U U
Figure 6.7. Block diagram of Lam's electronic portal imaging scanner. The
one-dimensional line of detectors move under motor control. The outputfrom
the detectors is amplijied and multiplexed, finally being displayed on a video
monitor. (From Lam et a1 (19861.)
ADC
TRIGGER
ANALOG
OUTPUT
Figure 6.8. Schematic of the signal-processing circuitryfor Lam's scanner.

+
The outputs from the diodes are held in a sample and hold ( S H) before
being multiplexed and output. (From Lam et a1 (1986).)
showed that, at doses comparable to those which would be received at treatment,

the contrast resolution (of 1 cm diameter disks) was better than 0.8% on a Varian
Clinac 4 / 1 0 4 MV accelerator.
The image was constructed throughout the full c o m e of a treatment fraction.
The accelerator was run at a pulse repetition frequency of 200 pulses per second.
A typical treatment fraction required 5000 pulses, and 20 pulses represented
0.33 cGy. The need to image throughout a treatment fraction is a severe
disadvantage for a scanning system of this type and demands that the quantum
efficiency of the detector be increased to completely offset the disadvantage of
viewing only a small part of the radiation field at any one moment in the scan.
Scanning silicon diodes have much the same quantum efficiency as a TV-based
system, but loose out because the latter is an area detector visualizing the whole
radiation field. A scanning system with some 50% efficiency has been developed

Scanning linear array of diodes and crystal detectors 257
Figure 6.9. Illustrates the arrangement of individual detectors in the Royal

Marsden Hospital portal scanner. Each crystal presents a square face (shown
hatched) to the radiation beam. One line of crystals is offset by half the square
size relative to the other. Samples are collected from both lines by integrating
for the time it takes one line to move halfa square. This arrangement satisfies
the Shannon sampling theorem. The arrow indicates the direction of scanning.
Only a few detecting elements are drawn.
by Swindell’s group (Morton 1988, Swindell et a1 1988, Morton and Swindell

1988, Morton et a1 1988, 1991b) making use of zinc tungstate detectors.
Lam et a1 (1986) also digitized simulator radiographs for comparison with
the digital images at treatment. They commented that portal images were
‘comparable’ in quality to film images, and more readily compared with simulator
images.
The system of Swindell at the Royal Marsden Hospital, Sutton (Morton 1988,
Morton and Swindelll988, Morton et a1 1988,1991b) has a scanning array of 128
zinc tungstate crystals arranged in two rows of 64, with the crystals of each row
offset by half a crystal size relative to the other row (figure 6.9). The crystals are
5 x 5 mm in area and 25 mm deep, arranged for practical reasons in 6 x 6 mm
spacing. Morton (1 988) determined that the light output of these crystals was some
twice that of bismuth germanate (BGO) crystals, resulting in some 2000 optical
photons reaching the photodiode per x-ray interaction, a favourable situation (see
section 6.1 1) for noise suppression. These crystals are non-hygroscopic, almost
resistant to radiation damage and have a high (2.2) optical refractive index.
The output from this continuously scanning detector was integrated for the time
taken to traverse half the detector spacing (i.e. for 3 mm). This constructed a 128
x 128 matrix image with the output from one row of 64 detectors filling the odd
columns and the output from the other row of detectors filling the even columns.
The data were low-pass filtered to remove machine variations. The detector was
2 m from the radiation source to reduce the contribution of scattered radiation to
the image, giving (with a magnification of 2) an image of size 19 x 19 cm at the
isocentre. The image took 3.8 s to construct, requiring a dose of some 0.55 cGy
(the dose to the patient was greater-25 cGy-because the beam irradiated the
full portal whilst the detector sampled only a part of the portal at any one time).
The experimentally determined spatial resolution was 2.5 to 3 mm and the density
resolution was O S % , in line with the design specification. This system (figure
6.10) is now being extensively used for clinical studies of the accuracy of patient

Figure 6.10. The Royal Marsden Hospital electronic portal imaging

instrument attached to a Philips SL25 accelerator. The scanner slots into
the triangular mounting plate, which is firmly bolted to the rotating gantry
of the linac. The acquisition electronics are located within the arm holding
the scanning detector. Thefigure shows the prototype detector; an improved
version is under development. (Courtesy of W Swindell, E Morton and P
Evans.)
positioning (Gildersleve et a1 1992~).Images produced by the portal scanner are

compared with digital images captured by the image intensifierof the radiotherapy
simulator or simulator-film digitizer (figure 6.1 1).
Before this comparison can be done, three spatial corrections are made to
the simulator images (Morton et a1 1991a) which were captured by a frame-
store coupled to a personal computer. (Incidentally, this provided the additional
advantage that workers in the simulator suite could view an image captured in
this way after the x-rays had been switched off, thus saving on patient and staff
radiation exposure). Firstly, the TV pixel aspect ratio of 4:3 is corrected to a
square image. Secondly, pin-cushion distortion is removed, observing that the
displacement of the image of a point from its true position can be expressed
as a Taylor series of the distance of that point from the origin, with just two
terms needed. Rudin et a1 (1991) have also reported distortion removal with a
polynomial fit. Thirdly, the simulator images were re-sampled to the same pixel
size as the megavoltageportal images. Tools for comparing simulator images with
megavoltage images (Evans et a1 1992a) and for correcting patient mispositioning
(Evans et a1 1992b)have been developed.
Gildersleve (1992) and Gildersleve et a1 (1992b,c) used the Royal Marsden
Hospital portal scanner to quantitate the random and systematic errors in patient
positioning during routine radiotherapy, retrospectively analysing data for brain
and pelvic irradiations. Being compact, the equipment was unobtrusive and could
be used for all but a few geometries without any danger of collisions. Random
error was defined to be the departure of the field positioning from the mean.

Matrix ionization-chamber detectors 259
Figure 6.11. ( a ) Portal image of a mantle field captured with the Royal
Marsden Hospital portal imager shown in figure 6.10. The lung shields
are seen as large white areas to the bortom left and right. The important
anatomical details (lungs, ribs and soft tissue) are clearly visible and the
images can be windowed as appropriate to visualize these more clearly on a
display monitor. (b)The corresponding simulator image. The image is made
up of two films to cover the large field. The location of the desired shielding
has been drawn by the clinician on to the films. (Courtesy of W Swindell, E
Morton and P Evans.)
Systematic error was the departure of this mean position from the ‘gold standard’
at simulation. For the brain, 95% of treatments had random positioning error less
than 3 mm. The mean systematic positioning error was about 1%,although (and
this was an important conclusion of the study) individual patients had systematic
errors as large as 6 mm. For the pelvis, 95% of treatments had random positioning
error less than 4.5 mm. The mean systematic positioning error was close to zero,
although individual patients had systematic errors as large as 6 mm. There was a
trend for patients with large systematic errors to also exhibit large random errors.
Whilst this study showed that positioning errors were within the margin which
many radiotherapists added ‘for safety’ in determining the target volume, the
existence of some ‘out-liers’ in the distribution emphasized the need to verify
patient positioning with portal imaging. Following this retrospective study, an
‘intervention’ study was commenced late in 1991 (Gildersleve et a1 1992a).
A scanning diode array has also been constructed at Massachusetts General
Hospital, comprising cadmium telluride detectors (Biggs 1991). The development
is at an early stage.
6.4. MATRIX IONIZATION-CHAMBERDETECTORS
Research at the Antoni van Leeuwenhoekhuisin Amsterdam has resulted in the

development of three prototype liquid-filled matrix ionization chambers for portal
imaging. The first was a small prototype of 322 chambers (Meertens et a1 1985)

and the clinical prototype had 1282chambers (van Herk and Meertens 1987,1988).
The present version has 2562 chambers (Meertens 1989, Meertens et a1 1990b).
The small prototype matrix ionization chamber of Meertens et a1 (1985)
comprised only 322 cells, giving a small field of view of 78 x 78 mm, operating
in much the same way as described below for the clinical prototype. The data
collecting electronics were, however, very slow, 120 s being needed for the
complete image. The small prototype allowed variable spacing between the
electrodes (1, 3 and 10 mm were tried) and could be filled with air or 2,2,4-
trimethylpentane. The liquid gave much greater cell current at each spacing than
the air, with little increase in the noise and was thus preferred (e.g. at a separation
of 1 mm the SNR was 200 (density resolution some 0.5%)with the liquid medium
and 15 with the air.) The smallest spacing was also preferred at a fixed voltage
of 300 V since, although more ions were produced with a wider gap, the lower
field strength allowed greater recombination, an effect dominating the increased
production. With a liquid-filled chamber operating at 300 V and a separation
of 1 mm, the spatial resolution was 3.8 mm with a 6oCo source. The spatial
resolution (measured by differentiating an edge-response function) was the same
in orthogonal directions and the point-spread function was deconvolved from the
measured images to improve their appearance. The point-spread function with the
liquid medium was narrower (at corresponding separations) for the liquid medium
than for the air. Hence 1 mm spacing with the liquid medium became the preferred
conditions for operation and were adopted for the larger clinical prototype.
In the clinical prototype the ionization chambers comprised a pair of 1.5 mm
thick glass-fibre printed circuit boards etched with a 2.5 mm pitch to give 128
rows on the front plate, serving as high-voltage electrodes, and 128 columns on
the rear plate, serving as low-current-collecting electrodes. Each crossing point
serves as an ionization chamber for measuring the local radiation intensity. The
electrodes were separated by a 1 mm film of the liquid iso-octane as ionization
medium. There was no additional radiation build-up material. Thus the field of
view at the detector was 320 x 320 mm. The ionization current in each chamber
was measured with a current-to-voltage converting electrometer. The ionization
matrix was scanned row by row, by switching the high voltage to different rows
sequentially and reading out the 128 column electrodes. The data were taken
through a 12 bit analogue-to-digital converter, stored on computer and the whole
image took 3.1 s to construct. Images could be displayed as a 256 x 384 array with
8-bit accuracy. This detector had the advantage of no moving parts, all ‘scanning’
being achieved electronically (figure 6.12). The advantage of a detector based on
ion transport with low mobility is that the ion lifetime is very long, resulting in
an inherent charge integration in the liquid. The Netherland’s group have also
considered a chamber based on electron transport (Boyer et al 1992).
The raw images required correction for electrometer offset (a variation of some
10%) and leakage current (a variation of some 1%). Corrections were also
made for changes in the ionization-chamber sensitivity (as great as 40%). This
arises because the response is very sensitive to the local thickness of the liquid

Matrix ionization-chamber detectors 26 1
Figure 6.12. Schematic of the liquid-ionization chamber megavoltage

imaging system. The camera cassette is connected with the control unit
(placed outside the treatment room) by a 15 m length of cable. The figure
shows the 1282 detector. (From van Herk and Meertens (1988).)
Figure 6.13. ( a )Raw image data of the RandolAlderson phantom imaged

with the 1282 matrix ionization-chamber portal imager in figure 6.12,
showing the electrometer offset and sensitivity variations, electrode shape
variations and variations due to the changing thickness of the chamber. (b)
The same image after rhe application of all correction steps. (From van Herk
and Meertens ( I 988).)
layer, which can vary with orientation of the detector on the gantry as deforming
pressures come into play. However, these corrections can be made during the
data capture itself. Figure 6.13 shows an image of a head before and after these
corrections.
Additionally (as with most megavoltage imaging systems) a number of fast
image processing algorithms were used, including linear contrast enhancement,
global histogram equalization, various 3 x 3 spatial filters, as well as image

Figure 6.14. Van Herk and Meertens’ electronic portal detector, as used to
verify the positioning of a patient for a radiotherapy treatment. (From van
Herk and Meertens (1988).)
arithmetic. All these operations took less than 0.6 s and were thus essentially
‘semi-real-time’.
Measurements were made (figure 6.14) using a Philips SL 75- 10 accelerator at
8 MV and showed a point-spread function with a full-width at half-maximum of
some 1.5 mm. Phantom measurements showed that the contrast was such that a
2.5 mm extra water thickness could be seen against a 200 mm uniform thickness
of water. The magnitude of the PSF was taken to indicate the absence of cross-talk
between ionization cells, despite the absence of chamber walls.
Van Herk and Meertens (1988) discuss the need for pattem recognition
software, so that the portal images which can be generated quickly can be used
to intervene in the treatment fraction. They also call for studies to evaluate the
use of electronic portal imagers in clinical practice and suggest they may also be
used as dosimeters. The authors of other systems make similarpoints. One special
claim for the matrix ionization chamber is that being ‘thin’ (the overall thickness
of the whole system is no more than 5 cm) it can easily take the place of a film
and be used with the same convenience.
The clinical prototype has been supercededby adetector with 2562chambers of
size 1.27 mm by 1.27 mm by 1 mm (Meertens eta1 1990b,van Herk 1991,1992).
The efficiencyis some 1%. A detailed analysis of the ion-chamber performance,
comparing with theoretical predictions, has been published by van Herk (1991).

Matrix ionization-chamber detectors 263
Figure 6.15. Portal images in the thoracic region made with the
high-resolution2562 imaging device during one treatment session: ( a )man-
tle field, (b)same image with additional high-passfiltering. (FromMeertens
et a1 ( I 990h).)(Reprintedwith permissionfrom Pergamon Press Ltd, Ogord,
UK.)
The 2562 detector could also be operated as a 1282detector with a lower noise
level. Figure 6.15 shows clinical images from this high-resolution device.
The Netherlands Cancer Institute system is available commercially as
Portalvision from Varian Inc, Palo Alto, Califomia (formerly the Dynaray ID from
Asea Brown Boveri). With the system 20 cm below the isocentre, the image field
is 27 x 27 cm2. Data may be acquired in either 1.5,2.8 or 5.5 s. In the commercial
system the fluid space is only 0.8 mm thick. The contrast resolution is quoted as
better than 0.5% (Varian 1992). The detector weighs 7 kg. Wade and Nicholas
(1991) report favourable use of the prototype system for a wide variety of patient
treatment sites. They note the need to recalibrate for oblique angles. The prototype
digital cassette was used free-standing,but the latest Varian version 3.1 is attached
to the treatment machine via a motorized support which enables a variable source-
to-detector distance, as well as lateral and longitudinal movements. This is also of
more rigid construction and does not need to be recalibrated at each gantry angle.
Meertens et a1 (1990a) describe methods for measuring field-placementerrors
using digital portal images. Bijhold et a1 (1991a) describe the edge-detection
algorithms developed for use with this system and Bijhold et a1 (1991b) show
how using features visible in simulator and portal images, assessment of patient
set-up error may be made. Bijhold et a1 (199 IC)report clinical usage of the system,
showing how intervention became possible. Bucciolini et a1 (1991) have used the
liquid-filled portal system (which was developed in Holland) in Florence.
Hoogervorst et a1 (1991) have reported a development of a 32 x 32 air-filled
ionization chamber, similar to the portal imager, used for measuring the flatness
of fields.

6.5. IMPROVEMENTS TO PORTAL IMAGING WITH FILM AS

DETECTOR
Three improvements to portal imaging have been made, retaining the use of film
as detector:
0 use of a film digiter,
0 use of a metal converter plate,
0 use of gamma multiplication.
6.5.1. Film digitization

A film digitization system was constructed by b o n g (1984). A trans-
illuminated film was viewed by a TV camera with the different optical densities
creating signals which were then digitized. The recorded digital images
were directly viewable on a monitor (Lexidata System 3400) attached to
the controlling computer (Vax 11/780) or were input to several digital-image
processing techniques and displayed to show a modified or enhanced version of
the image. Contrast re-scaling, histogram equalization and filters applied in the
Fourier domain were implemented. It was claimed that the resulting images were
more easier to interpret than peering at low-contrast radiographs on a light box
showing information present but not easily perceived with the naked eye.
The images were corrected for the non-uniform illumination of the light box; no
spatial distortion was observed. The spatial resolution was limited by the discrete
sampling which, in principle, leads to aliasing, since (with such a small grain size)
the image is essentially not band-limited. However, it was thought that all the
useful information was within the band limit of the digitization (the images were
described as ‘pseudo-band-limited’) and that any aliasing would only introduce
noise which was removed by the Fourier filters.
The system was shown to have a useful dynamic range of about 80, not a
limitation because of the inherent lack of contrast in portal films. When 16 or more
frames were averaged, the signal-to-noise ratio (defined as the ratio of the mean to
the standard deviation in the response to a uniform field) was about 60, indicating
that 6 bits only were required to digitize the image. The principal source of noise
was thermal noise in the TV pre-amplifier; film-grain noise and photodetector
noise could be ignored.
It is thus clear that there is a much smaller dynamic range and signal-to-noise
ratio in digitized portal films than in some on-line electronic portal imagers.
Meertens (1985) digitized film differently using a densitometer with a spatial
resolution of 1 mm and density resolution of 0.2% displaying images as a 2562
pixel matrix on the console of a CT scanner. The films had been exposed
sandwiched between two copper screens of thickness 0.6 mm (front) and 0.4 mm
(rear). Thus he brought into play all the computational facilities available for
inspecting CT tomograms, such as grey-scale windowing, filtering, distance
measurement and inspection of regions of interest. A method was also developed

Improvements to portal imaging with film as detector 265
to superimpose data from the simulator film for comparison purposes. The
main flaw with these procedures was the time factor, too long to allow real-time
intervention at the point of treatment, a limitation noted by all workers (including
Meertens) proposing real-time portal imaging.
Digitization of port films has been widely applied. Amok and Lowinger (1987)
used a black-and-white video camera to digitize to 5122, 8 bits deep, whilst
Grimm et a1 (1987) used a Dupont laser scanner to create images at 1024*. In
both cases the digitized images were further processed by contrast enhancement,
windowing, edge enhancement and filtering to render structures of interest more
clearly. Grimm et a1 (1 987) reported significant improvements in head and neck
imaging. Crooks and Fallone (1991) built a system around a personal computer.
6.5.2. Influence of metal screens

Metal screens are generally used with film. Without the screen both primary and
scattered photons largely pass undetected through the film emulsion, so the film
can only respond to electrons generated in the patient and specifically in the last
few cm of the patient (since electrons of N MeV travel only some N/2 cm in
tissue). This is undesirable because of the long range of such electrons. If there is
an air gap between the exit surface of the patient and the detector, the image will be
swamped by the electrons from scattered photons and will be blurred. Even those
electrons resulting from primary interactions will not deposit their energy at the
correct location because of the finite air space across which they are transported.
The purpose of the metal screen is to totally remove the contribution of electrons
from the patient (provided the screen is thicker than the maximum electron range
in the metal-some N /22 cm for a lead screen) and instead generate new electrons
by photon interactions in the lead. These travel a distance of some 11 times less
than they would in tissue and so are sharply imaged. The screen provides aflat
build-up layer directly next to the film (figure 6.16). The electron production in
the screen reflects the scatter-to-primary ratio of photon fluxes, provided both have
built up to electronic equilibrium. The front screen does not act as an intensifier
since it does not provide any extra electronic contribution to the film. It simply
replaces the patient-generated electrons by a new set generated in the film and
more localized to the points of interaction. (By contrast a rear screen does have
an intensifying effect, since it provides backscattered electrons to increase the
electronic contribution to the film-see later; this effect is by a factor of about
1.8.)
The effect on the image contrast of screens of different material and thickness
can easily be related to their effect on the scatter-to-primary ratio of dose at the
film. Droege and Bjamgard (1979) studied this experimentally with some very
clear conclusions. Suppose two parts of a film receive primary doses PI and P2
and the same scatter dose S, then the dose contrast is

I \
e-
h
Figure 6.16. Showing the effect of a metal screen in front of a portal film.
The patient is shown in cross section and the fan lines represent a beam
irradiating the target area shown. In the left-hand figure, both photons
(primary and scattered) and electrons exit the patient. The electrons have
come from the lastfew cm of the patient nearest the film (shown dotted). Most
photons pass straight through the film, which responds to the electrons. In
the right-hand figure, the same particles exit the patient. The electrons from
the patient are stopped by the metal screen, in which the photons interact to
produce a new set of electrons to which the film responds. This is a more
favourable situation.
Writing A P = P2 - PI and P = PI we have
c=1+-
Y -
( 1 +1s/P).
This gives a film-density contrast
AD=ylog(C). (6.5)
where y is the slope of the Hurter-Driffield curve.

It therefore follows that the only factors which can affect the contrast in
density (which is the important physical measure of whether a small object can be
perceived) are y and the scatter-to-primary ratio S/ P (figure 6.17). Droege and
Bjamgard (1979) found experimentally that whatever the source energy (%o, 4
MV and 8 MV), whatever the screen metal and whatever its thickness, the Hurter-
Driffield curve of film density versus film dose (corrected for the transmission
factor of the screen) was the same and so the film gamma was unaffected by the
presence of the screen (figure 6.18). This simply reflects that the film responds
to electrons generated in the screen and, provided these are in equilibrium, it is
irrelevant what material they come from or how thick it is.
To consider further image contrast on portal films, attention therefore turns
to the S I P ratio. This was investigated for contact geometry (screen and film
touching the exit surface of a phantom) and for a geometry with an air gap (screen
and film some 25 cm from the exit surface of the phantom). In contact geometry

tog FILM msE

Figure 6.17. Illustrating thefactors which can affect radiographic contrast.
These factors are the film y and the scatter-to-primary ratio. (From Droege
and Bjarngard (1979j.j
I KODAK XM FILM ~ P
_
Energy _
Screen - ~
g cm-2
6%o None
6OcO pb $15
4MV None
4MV Pb 014
4MV Cu OC9
BMV None
BMV pb 014
BMV Ft 231
BMV CU 009
8MV Cu 295
4, ' ' " 0" "5' !

I ,
5
I ,
REi A T/ VE FfL M DOSE
Figure 6.18. Showing the measurements which indicate that the film
contrast is independent of x-ray energy and metal screen composition. All the
measurements lie on a single curve. (From Droege and Bjarngard (1979).)
the S / P ratio initially increased as the thickness of the screen is increased, but
after a certain thickness settled to the same value as without the screen at all. This
optimum thickness is 1.5 g cm-2 for copper and 2.5 g cm-2 for lead, the result
largely independent of beam energy (figure 6.19). In a geometry with an air gap,
the S I P ratio decreased monotonically as the screen thickness is increased, from
a high value with no screen thickness to a roughly constant value after 1.5 g cm-2
of either material. The use of an air gap decreased the S / P ratio considerably
from the corresponding value in contact geometry (as also shown by Swindell et
a1 1991), thus improving contrast (excepting the extra blurring introduced by finite
source size).

L+--+lL
OL 0
A pb t 0.43 q.Cm"Al
THfCKN€S, g cms2
Figure 6.19. Showing the effect on the scatter-to-primaryratio S/ P o f f r n t

metal screens of different thicknesses and two materials used in megavoltage
contact radiography at 6oCo energy. When the thickness reaches a certain
value, the S/P ratio is the same as when there is no metal at all. This is the
best thickness to use. (From Droege and Bjarngard (1979).)
To understand the behaviour of the S/ P curve for contact geometry as a function

of screen thickness, it is necessary to consider carefully the photon-electron
transport in phantom, screen and film. Firstly, with no screen the dose to the film
is provided almost entirely by the electrons which are generated from the layer of
phantom or patient nearest the film; the film itself stops negligiblephotons at these
energies. Hence the scatter-to-primary ratio simply reflects the ratio of electrons
from scattered-to-primaryphotons at the exit of the phantom, which in turn mirrors
the ratio of scattered-to-primaryphotons at the exit surface. Droege and Bjarngard
(1979) checked this was so using published depth-dose data.
A thin screen has a dramatic effect on the mostly forward-travelling electrons
liberated in the phantom by primary photons, turning a large fraction away from
the film. As the thickness of the screen is increased, both scattered and primary
photons liberate electrons in the screen, the build-up depth being smaller for
scatter than for primary. Once electron equilibrium is achieved, the film dose
becomes proportional to the screen dose from such photons. However, since the
energetics are different, these doses will decrease at a different rate with increasing
screen thickness. The overall result is the form of the curve shown in figure 6.19.
Although the thick screen produces almost exactly the same scatter-to-primary
ratio as no screen, it is preferred because of the resolution-degrading effect of
long-range electrons generated in the patient or phantom, all of which have been
excluded from the film by the screen (figure 6.16). In summary the main purpose
of the screen is to remove patient-generated electrons, replacing them by a new set
generated in the screen and more localized to the points of interaction. The scatter-
to-primary ratio of a screen of build-up thickness or greater has to be the same as
with no screen, because one can think of a lead screen as acting just like 11 times

its thickness of 'extra patient' as far as producing electronic fluxes proportional to

the two photon fluxes; primary and scatter. All that changes is the location of the
production sites in relation to the film (i.e. right next to it).
If we write
A P I P = ew' - 1 N pt (6.6)
for the ratio of the change in primary dose to the primary dose from a small
extra attenuating thickness t of water-equivalent material of linear attenuation
coefficient p and put this in equations (6.3) and (6.5), we find the minimum
perceptible tissue-step thickness tmingiven by
tmin= ln(l0)-
ADmin (1
Y
- +iff')
Using Droege and Bjamgard's data for S / P and the known values of p at the three
energies studied, it tums out that tminis remarkably the same at all three energies.
Yet it is known that the contrast in portal images with 6oCoradiation is worse than
with 8 MV; the difference is attributed to changes in spatial resolution.
Galkin et a1 (1978) also investigated the dependence of film y on the presence
of metal screens. From measurements with Kodak Rapid-Process therapy
film (RP/TL) and the radiations from 6oCo (Picker C-3000), 4 MV (Clinac-
4 accelerator) and 45 MV Betatron (Brown Boveri), they determined that the
presence of a single lead screen made no difference to the film gamma, in
agreement with Droege and Bjamgard (1979). Sandwiching the film betwen two
screens made small increases in y . However, using just a single light-emitting
screen (US Radium Radelin Ultra-detail screen) increased the film y considerably,
an effect even greater with a pair of light-emitting screens. This is because the film
responds differently to electrons (from the metal screen) and to optical photons
(from the light-intensifying screen).
In Sweden, Jevbratt et a1 (1971) used low-sensitivity industrial film with a y of
6.5 to record portal images from 6oCoradiation, a 6 MeV linear accelerator and
a 42 MeV betatron, improving the sharpness by employing a sandwich of 2 mm
thick lead screens to remove the scattered electrons from the patient and replace
them by a new set of electrons generated in the screens. They also investigated
graphic film with a y of 12 and very slim latitude.
Hammoudah and Henschke (1977) showed experimentally that the effect of
a front lead screen was to decrease the effect of scattered electrons and thereby
increase contrast with no intensifying effect, whereas the rear screen produced an
intensifying effect. A regular diagnostic film sandwiched between two lead layers
was some ten times more sensitive than supervoltage localization films. They
found films used with fluorescent screens were too sensitive, being over-exposed
at the minimum monitor setting. Similar observations on the different physical
effects of front and back metal intensifying screens were made by Halmshaw
(1966) in his large text on industrial radiography.

6.5.3. Film enhancement by analogue processing

Instead of numerical methods for enhancing digitized films, an analogue technique
using purely photographic processes has been developed by Reinstein and Orton
(1978,1979). This method is called ‘gamma multiplication’ for reasons which
will soon be obvious. The original poor-contrast portal radiograph is copied by a
contact process onto a second previously unexposed film. This print has reversed
contrast, of course, being exposed more where the density of the portal film was
less. The contrast on this reversal film is the product of the gamma of the portal
film and of this copy film. Next, the process is repeated by making a contact
print of the second film onto a third film, at which stage an image with the same
polarity as the original is recovered. The contrast on this third film is the product
of all three gammas of the individual films. If the first is a Kodak XV with y =
2.3, the second is Kodak XTL with y = 3.5 and the third is also Kodak XTL, the
gamma multiplication is 28.
The method is rapid and inexpensive. Good contact is needed between films and
this is ensured by placing both under a heavy glass screen. The contact prints are
made by normal darkroom techniques and require no special apparatus. Reinstein
and Orton (1978, 1979) verified the gamma multiplication experimentally by
imaging a thin step wedge, providing initially very small contrasts, and observing
the contrast magnification at each stage.
There are trade-offs, of course. Increased contrast comes at the expense of
decreased latitude. Indeed, contrast enhancement may be so great with two XTL
films that a third-stage film with lower gamma may be preferable. A second
negligible disadvantage is that copying decreases spatial resolution, but this is
not a limiting factor with portal radiographs at high energy.
With reference to figure 6.20, the gamma multiplication may be shown
mathematically: Figure 6.20(a) shows the formation of the original film with an
incident intensity of radiation X O producing primary output intensities X I and XZ
corresponding to passage through two regions in the brain with contrast. The
corresponding film densities are D 1and D2.By definition of the film gamma,
Di - Dz
Y1 = log x1 - log x2 ’
Figure 6.20(b) shows the first copying stage. Light of intensity Zo is shone onto
the film giving the two light output intensities ZI and Z2, which create the new
densities D{and D;on the copy film. The copy-film gamma y2 is then given by
Di - 0;
Y2 =
log I1 - log 12 *
Since, by definition of optical density D I= log ZO - log ZI,this reduces to
D;- 0;
Y2 = (6.10)
D2 - D I

Imaging with a photostimulable phosphor plate (Fuji system) 271
which, by equation (6.8), gives
(6.1 1)
Since (0; - D;)/(log X1 - log X2) is the definition of the combined gamma,
which we can write y1,2,we have
Y1,2 = -YIY2* (6.12)
This shows the gamma multiplication at the first stage. The minus sign shows
the reversal process, The argument is repeated at the third stage. We have (see
figure 6.20(c)) light of intensity Zh irradiating the copy film and giving rise to two
intensities Zi and I;, which in tum give densities 0" and 0;' on the third film.
Again by definition-
0" - 0;'
Y3 = (6.13)
log z; - log 1;
and, from the definition of optical density (by analogy with equations (6.9) and
(6. IO) above),
0; - 0;'
Y3 = (6.14)
0; - 0; *
Since (0;- D;)/(log X1- log XZ)= -yl yz, we have
0; - 0;
Y3YlYZ = (6.15)
log x1 - log x2
the right-hand side of which is the overall effective gamma, ye^, and so finally
The phase reversal is reversed back to a positive image and this proves the 'gamma
multiplication'.
6.6. IMAGING WITH A PHOTOSTIMULABLE PHOSPHOR PLATE (FUJI

SYSTEM)
Computed radiographic imaging has been developed by the Fuji Photo Film CO
(Sonoda et a1 1983) for imaging at diagnostic radiology energies. The same
equipment has been used for megavoltage portal imaging by Wilenzick et a1
(1987). The image receptor comprises a flexible plate, some 1 mm thick, coated
with europium-activated barium fluorohalide compounds in crystal form in an
organic binder. The photostimulable phosphors act as energy traps when exposed
to ionizing radiation, producing a stored or latent image. When illuminated by red

Fz luiexpoisdl Fz IreversoI)
I i I
X-Omot Processor
~ c) NNAL ENHANCEMENT
-2 F3 hexposed) F3
I I
X-Ornot Processor
Figure 6.20. The method of 'gamma multiplication'. This is a three-stage

process: ( a ) showing the production of a portal radiographic film, (b)
showing the production of an enhancement reversal film,( c ) showing the
production of the third and final enhancement film. The symbols are defined
in the text. (From Reinstein and Orton (1979).)
laser light at 633 nm, the stored energy is emitted as light. The usual technique
to do this is to shine the laser light onto a scanning mirror, which illuminates
one small part of the plate, the emitted light being taken into a light guide to a
photomultiplier tube and thence converted to electrical energy.
The imaging sequence (figure 6.21) starts by releasing any traps by flooding
the plate with light. After exposure to radiation, an initial low-resolution (2562)
image is 'pre-read' with a low laser intensity to set the sensitivity. Then the image
is read at high resolution (2O0O2)with 8 bits per pixel at a higher laser power. The
images can be stored as digital computer files and processed in the usual ways, for
example by windowing, contrast adjustment (by simulating some Hurter-Driffield
response curve), by unsharp masking, etc.
This imaging modality is sensitive over four decades of response, exhibiting a
linear dose-photoluminescent response function. For this reason it can cope with
the overexposure by some two orders of magnitude compared with how it is used
at diagnostic energies.
Wilenzick and Merritt (1987) used the system to record images at 6 and 10 MV
and showed, using the Lutz-Bjamgard portal-film test object (Lutz and Bjamgard
1985), that the computed radiography system outperformed conventional portal
imaging with film. They discussed the interesting possibility of using the portal

Portal imaging by reconstructing from projections 273
Photostimulable phosphor crystals

support
X-ray quanta
Scanning mirror
n
Exposure
Reading
n
Imaging plate
Figure 6.21. The imaging method with a photostimulable phosphor screen

showing exposure of the plate, release of latent energy by scanning laser, and
erasure of the image to prepare the plate for re-use. The laser-stimulated
fluorescence is detected by a photomultiplier tube from which the digitized
signal is obtained. The mirror is used to scan the laser beam over the plate.
This technique is also used in diagnostic radiology. (From Wilenzick and
Merritt ( I 987).)
data for transit dosimetry (see section 6.12).

A commercial version of the portal imager, based on the use of a
photostimulable phosphor, is manufactured by Siemens (the 'FCR 7000
Digiscan'). This can give five image formats with pixel sizes varying between 100
and 200 km, giving a spatial resolution of between 2.5 and 5 lp mm-' (Scheck et
a1 1992). The images are impressive, but of course this is not a 'real-time' method
of megavoltage imaging.
6.7. PORTAL IMAGING BY RECONSTRUCTING FROM PROJECTIONS
Bova et a1 (1987) described a novel detector comprising a rotating and translating

one-dimensional, strip, ion chamber, collecting projection data from which the
portal image could be reconstructed by filtered backprojection. Several types of
ion chamber were tested, with a liquid-filled chamber emerging as favourite. The
images are digital and exhibit good scatter rejection. A spatial resolution of 4.1
mm was achieved when reconstructing from 180 angular samples (Boyer et a1
1992) (figure 6.22).
Now one may imagine a set of long, thin ionization chambers positioned parallel
to each other so as to form a detection plane. At one orientation of the detector a
1D projection of the portal image is recorded. Now imagine the detector rotates,
collecting a 1D projection at each of a large number of orientations in 180". The
2D portal image may now be reconstructed from these 1D projections. There is no
need for the detector to translate. This is the same principle as engineered by Keyes
(1975) and by Webb er a1 (1992) to make a novel gamma-camera collimator which
improves on the performance of a matrix of parallel holes with direct capture of

Figure 6.22. Illustrating the principles of how a 2 0 image can be made

from a one-dimensional detector D which integrates all the signal along its
length. Imagine this detector translates in the direction of the arrow without
rotating across the field F to be imaged; a one-dimensional projection of the
2D image would be formed. Now if this scanning movement is repeated at a
set of orientations over 180°, the 2D image could be reconstructedfrom the
set of projections.
Figure 6.23. A rotating parallel-walled slit collimatorfor a gamma camera.

The collimator rotates through 180°,collecting a projection of the ZD planar
image at each orientation from which the 2 0 image may be reconstructed.
the 2D image (figure 6.23). The equivalent megavoltage imaging system has not
been engineered.
6.8. PORTAL IMAGING BY XERORADIOGRAPHY
Another diagnostic imaging technique which has been pressed into use for high-
energy portal imaging is xeroradiography (Fingerhut and Fountinell 1974, Wolfe
et a1 1973). Xeroradiography dates back to the 1890’s and was extensively
developed for diagnostic mammography from the 1960’s (Boag 1973). The
same commercial equipment has been used for megavoltage imaging. The
xeroradiographicimaging process produces images with strong edge enhancement
and little broad-area contrast (Dance 1988).
The xeroradiographic process exploits the photoconductive properties of
amorphous selenium. The receptor is a thin layer of selenium (125pm) deposited

Solid-state imagers 275
on an aluminium backing plate. The detector is charged prior to irradiation.

During irradiation the energy absorbed creates electrons and holes. These charge
carriers migrate to the surface of the selenium under the action of the electric
field, forming a latent image by subtraction from the original uniform charge
distribution. The charge image is developed by being sprayed with a fine aerosol
of blue powder and a powder pattem builds, reflecting the charge image. The
powder pattem is transferred to paper by contact and fused by heating.
Fingerhut and Fountinell (1974) used xeroradiography to make images using a
4 MeV accelerator with a dose of 4 cGy. The wide latitude of the process allowed
good visualization of anatomical detail. No details were published. Development
took 90 s and they claim the method completely replaced the use of film at their
clinic. It might be observed, however, that the method has not really caught on
elsewhere. In principle a digital image could be produced.
At much the same time, Wolfe er a1 (1973) used xeroradiography to capture
portal images from a 6 0 ~ source,
o presenting images of a variety of treatment
sites in the body, showing the versatility of xeroradiography consequent on its
wide latitude. For example, air spaces and bones were simultaneously visualized
in one single exposure (radiation dose 6 to 9 cGy) of a large treatment field (for
bronchogenic carcinoma) designed to include regional lymph nodes. Wolfe er a1
(1973) claimed the method was ‘quick and convenient’;they quoted times of some
2 minutes for development, claiming this was sufficiently short for the therapist
to verify the accuracy of positioning the radiation field and readjust if necessary,
repeating the imaging. Today this statement would find opponents amongst those
striving for near-real-time portal imaging for truly interventional studies.
6.9. SOLID-STATE IMAGERS
Antonuk et a1 (1990a,b, 1991a,b,c)have reported the development of a solid-state

detector consisting of amorphous silicon photodiodes and thin-film transistors. A
prototype detector of 11.5 cm by 10.8 cm was constructed with a 450 pm pixel
pitch. This is a very new development; the eventual aim is a 30 cm by 30 cm
system or even bigger.
The detector comprises a matrix of hydrogenated amorphous silicon (a-Si:H)
photodiodes each coupled to a single a-Si:H field-effect transistor (FET). This
matrix is constructed by techniques which are similar to photolithographic and
etching methods used to create conventional crystalline silicon circuits. The
silicon detector is typically 1 to 2 p m thick, deposited on a chrome metal substrate,
comprises three layers of a-Si:H with different doping, capped by an indium
tin oxide conducting layer. X-ray photons are converted to optical photons
+
by a suitable metal phosphor screeen. Antonuk et a1 (1991a) have studied
the effects on signal and noise of different screens with different thicknesses,
together with the variations with thickness of the a-Si:H detector. Other important

considerations are the effect of the ‘dead space’ taken up by the data lines, FET
lines and the bias lines.
The development is known as ‘MASDA’, multi-element amorphous silicon
detector array. The potential benefit for radiotherapy portal imaging is the
compactness of the detector; comparable to a film cassette. The major advantage
of MASDA is that, although the optical photon production efficiency is much the
same as for all systems using a metal plate and a phosphor, the detecting plane is
right next to the source of optical photons, resulting in very small losses. Hence
<
some 50% of the light signal can be measured, unlike the 1% for a fluoroscopic
camera or fibre-optic system (Boyer et a1 1992). Amorphous silicon materials are
also inherently radiation resistant. Research is progressing towards a detector of
comparable dimensions to a film; very recently the first radiotherapy images have
been obtained (Antonuk et a1 1992).
6.10. DIAGNOSTIC IMAGING ON A LINEAR ACCELERATOR
There have been attempts to take diagnostic-energyradiographs with a diagnostic

x-ray tube attached to a high-energy linear accelerator. Since these are not true
‘megavoltage images’ they are included here more for completeness. Diagnostic-
quality x-ray films were obtained many years ago by mounting an angularly offset
x-ray source and detector on the side of a cobalt treatment unit (Holloway 1958,
Shorvon et a1 1966). Ideally the diagnostic energy source should coincide exactly
in position with the high-energy treatment source and, according to Biggs et a1
(1985), this has been engineered at Stanford University by Varian Associates.
However, the next-best solution is to mount an x-ray source on the therapy
machine, offset from the treatment head by a precise angle (figure 6.24). Biggs et
a1 (1985) report a development of this type at Boston. The important feature of the
development was that the diagnostic imaging assembly included an exact replica
of the support system for holding the blocking tray. The axes of the diagnostic
and therapy beams coincided at the isocentre of the machine and were 45” apart.
The low- and high- energy x-ray sources were exactly the same distance from the
isocentre. The extra weight of the diagnostic tube was counterbalanced. Great
care was taken to provide devices for precisely engineering the position of the
diagnostic x-ray tube, and which allowed a good programme of quality assurance
thereafter. A ‘two-bubble’ spirit-level-like device was used when moving from
the diagnostic-imagingmode to the therapy mode. Two spirit levels were mounted
on the gantry at the same angular separation as the two x-ray axes, and movement
between the two modes was such that one or other bubble was centred for each
mode. This ensured that the diagnostic-energytube during imaging was exactly
in the same place as occupied by the therapy source during therapy. Possibly the
reason these ideas were not taken further is that they removed the ability to rotate
the collimator (i.e. to ‘put on head twist’).

Theoretical considerations of dose and image signal-to-noise ratio 277
Figure 6.24. View of a diagnostic x-ray unit attached to the head of a linear
accelerator. The central axis of the diagnostic beam intersects that of the
therapy beam at the isocentre and the two lines are at a jixed orientation
to each other whatever the gantry angle. There is a complete replica of
the accessoriesfound on the linac (e.g. the wedge tray) incorporated into
the diagnostic x-ray unit. This arrangement allows ‘diagnostic-quality’
radiographs to be obtained with the patient in the treatment position. (From
Biggs et a1 (1989.) (Reprinted with permission from Pergamon Press Ltd,
Ogord, UK.)
6.11. THEORETICAL CONSIDERATIONS OF DOSE AND IMAGE

SIGNAL-TO-NOISE RATIO
Viewed as a radiological imaging system, albeit at an unusually high energy, many

of the phenomena, well understood in relation to diagnostic-energy x-radiology,
can be investigated by similar mathematical arguments. Surprisingly, this had not
been done in detail until the recent paper by Swindell et a1 (1991).
They established that the minimum dose Dmi, at the build-up depth dm,,
required to see an object of area l 2 and thickness x in a slab of tissue of thickness
T and linear attenuation coefficient p with monoenergetic photons of energy E,, a
detector of quantum efficiency q , a scatter-to-primary ratio SPR with a differential
signal-to-noise ratio DSN R , was
Dmin = E p (%)m
DSNR~ L1
+ )2 (1
( L 1 - t dmax
+ f) f (1 + S P R )
(6.17)
where L1 is the distance from the source to the centre of the ‘object’ to be viewed,
which is itself a distance t below the surface of the slab, and A p is the change in
linear attenuation coefficientbetween the object and its surroundings (figure 6.25).
p e n / P is the mass-energy absorption coefficient of the soft tissue. The factor g
accounts for the Poisson noise in the random detection process (it is the same

SOURCE
Figure 6.25. The geometry used to evaluate equation (6.17).A tissue slab
of thickness T contains a piece of bone of thickness x at a depth t below
the sulface. Two lines of radiation are shown, one passing through the bone
onto a detector element DE T2 and the other passing through just soft tissue
to a detector element DE T I . The bone and tissue have linear attenuation
coeficients 4*2 and ~ 1 respectively.
, (From Swindell et a1 (1 991).)
as the factor P3 P4g2 in section 6.2). The factor 1

9) (+
is almost exactly unity
for scintillation-detection portal imagers, but can be as large as three for TV-based
systems. Swindell et a1 (199 1) present detailed calculations of the SPR in different
geometries which, together with the nomogram (figure 6.26) for bone ‘object’ in
soft tissue (derived from this equation when S P R = 0), can be used to predict the
dose-size trade-off under various circumstances. Even small air gaps (say 40 cm)
were found to reduce the SPR to less than 10%. Following Rose’s theory (Rose
1948,1973, Webb and Johnson 1990) the D S N R was set to 10 (i.e. aconservative
estimate of double the usual Rose factor of 5 for the smallest DSN R that can be
perceived).
The ‘diagnostic equivalent’ of the above equation was derived by Dance (1988).
When comparing TV-based systems with scanning-detector arrays, the different
quantum efficiency of the two systems must be considered, as well as the obvious
fact that one system involves scanning and the other does not. Because the x-ray
beam cannot be steered in synchrony, the dose to the patient is higher than the
dose forming the image with a scanning system.
The number of bits required to represent the image faithfully works out to be
about a minimum of 10 bits in typical conditions.
6.12. PORTAL DOSE IMAGES; TRANSIT DOSIMETRY
Portal images, whether accrued in analogue fashion on film or by electronic

imagers, have to date largely been correlated with simulator films, or radiographs
digitally reconstructed from CT (DRR) to assess position uncertainty. The use of
DRR is not yet well established, perhaps because of the limited spatial resolution

Portal dose images; transit dosimetry 219
T : 250 mm (patient
5SNR = I O thickness]
100 2 3 4 5 6 7 8910’
I (mml
Figure 6.26. Nomogram showing the minimum dose required to see a bone
of square shape ( I x I mm with CL = 0.0083 mm-‘ embedded in the surface
( t = 0 ) of a 25 cm thickness of tissue (CL = 0.0049 mm-’). The curves are
parametrized in terms of the bone insert thickness, x. The effect of scatter is
not included here, as equation (6.1 7j shows how the nomogram changes with
different scatter-to-primary ratios and these can be worked out from Monte
Carlo calculations for different geometries. (From Swindell et a1 (1991).)
compared with simulator films, arising because of the finite slice width of the
CT data from which they are generated (Sontag and Purdy 1991). Whilst these
comparisons are valuable and can give indications of patient mispositioning, they
cannot, for example, determine whether a wedge has been inadvertently inverted
or an incorrect wedge used. Ideally the comparison should be between measured
portal dose and the predicted portal dose.
Measurements of portal dose are called portal dose images or PDIs. Portal dose
can be predicted if a 3D CT data-set is known and the model for calculating the
exit portal dose is accurate. The latter must have a full description of scattered
radiation. It is not adequate to simply ray-trace the primary radiation through the
patient.
Wong et a1 (1990b) have compared the predictions of the ‘Delta Volume’
calculation method with measurements using TLD chips and an ionization
chamber for a phantom study of known composition and for a Rando phantom
where CT data were available. They report that for these well controlled conditions
agreement was better than 1%. The same could not be said for an attempt to match
a PDI for a patient with a calculation, when the portal dose was found for a small
number of spot measurements. Errors as large as 10% arose. However, when a
film was digitized to make an areal PDI, more informative results were obtained.
Wong et a1 (1990b) were able to demonstrate that adjusting the position of the
patient could bring the calculated PDI into good alignment with the measured PDI.
This is a very difficult area of research and there is still much to be done (Swindell
1992a).
Ying et a1 (1990) described a very complicated way of using portal dose images.
A portal dose image was calculated from 3D CT data and compared with the

measured PDI. Then the ratio of the two measurements was distributed back
along fan lines through the CT volume to adjust the CT data. From these data,
a new calculation of PDI was made and the process cycled until the calculated and
measured PDI were close enough in agreement. At that stage the adjusted CT data
were used to compute patient dose. All this took place of course after the patient
had been treated, since the forward dose calculations were lengthy.
6.13. MEGAVOLTAGE COMPUTED TOMOGRAPHY
In Tucson, Arizona, and later at the Royal Marsden Hospital, Sutton, Swindell and
colleagues have developed megavoltage computed tomography (MVCT) (Simpson
et a1 1982, Swindell et a1 1983, Lewis and Swindell 1987, Lewis et a1 1988,1992).
A Swedish development has also been reported (Brahme et a1 1987, Kallman et a1
1989) as well as a replication of Swindell's system in Japan (Nakagawa et a1 1991).
Megavoltage CT scanning has been reported in the literature of non-destructive
testing, but is rare with medical applications (Kanamori et a1 1989).
Strictly MVCT does not fit into the scope of a review of portal imaging, but
is included here because the work is complementary. The developments achieve
several goals. Firstly, it may be noted that x-ray linear attenuation coefficients
derived from a machine operating at megavoltage energies would be immediately
applicable to making tissue inhomogeneity corrections in planning (whereas for
a diagnostic machine a conversion is needed). Secondly, images taken on a linac
would help to ensure the patient was in the treatment position. The aim is to
take images just prior to treatment to verify the positioning of the patient by
comparison with CT images at treatment simulation. This provides a connection
with CT imaging on a simulator, since these latter images might well be obtained
for certain treatments (e.g. breast radiotherapy) with a simulator non-diagnostic
computed tomography (NDCT) scanner (Webb 1990). Periodic checks of the
patient anatomy (to determine, for example, whether a tumour has changed its
size) become feasible during the course of fractionated radiotherapy.
Swindell's work has progressed in several phases. Using the well known
relationship between image spatial and density resolutions, slice width, beam
energy, delivered dose, incident x-ray energy, detective quantum efficiency and
object size (Barrett et a1 1976), they calculated that, by relaxing the spatial
resolution requirement to some 2 mm, images with a signal-to-noise ratio greater
than 200 could be calculated with a dose of about 8 cGy. When the signal-to-
noise ratio was relaxed to 100 the dose could be as low as 2 cGy. The first system
constructed had a multi-element detector made of Pilot-B plastic scintillator
coupled to photodiodes. The 80 individual elements were arranged on the arc of a
circle of radius 140 cm centred on the x-ray source. 110 fan-beam projections
(12 bits deep) were collected in a 220" arc and images were reconstructed by
a convolution and backprojection (CBP) method with a SheppLogan filter into
128*matrices using a PDP 11/34 computer. Simpson et a1 (1982) showed that the

Summary 281
relationship between reconstructed CT numbers and tissue electron density was

precisely linear. The measured resolution was some 4 mm and a signal-to-noise
ratio greater than 100 was achieved.
The plastic scintillator was soon abandoned in favour of a detector fabricated
from 96 bismuth germanate crystals. These had a superior physical density (7.13
g ~ m - instead
~, of 1.1 g cm-3 for the plastic) and a stopping power of 83%instead
of 25%. The distance of the detector from the source was also increased from 140
cm to 180 cm, decreasing the scatter contribution to the data. The reconstructions
showed once again a linearity of better than 1 % between CT number and electron
density, a spatial resolution of 3 mm and signal-to-noise ratio in excess of 100.
The dose delivered was some 10 cGy and indicated that there was still room for
improvement in the data gathering and handling. Also, annoying circular artefacts
were seen, caused by electronic drift. The data were, however, clearly adequate
for treatment planning and verification purposes and Swindell et a1 (1983) showed
a variety of in vivo images, mainly of head structure in view of the relatively small
field of view (28-32 cm). They concluded that the limit on spatial resolution in
such a system would probably be set by the finite spot size of the x-ray source and
they felt it was still an open question as to whether megavoltage CT would allow
tumour visualization.
Lewis and Swindell (1987) and Lewis et a1 (1988, 1992) report setting up a
very similar system to the second system (above) at the Royal Marsden Hospital,
Sutton (figure 6.27). The major improvements over the Tucson system included:
(i) image formation with less dose (5 cGy) but with no reduction of image quality,
(ii) image reconstruction during little more than the time it takes to scan, (iii)
shorter reconstruction time, (iv) the detector takes just a few minutes to attach
instead of the half an hour for the prototype, making working with the system
much more convenient, (v) the system has been put into clinical use. An image
from this system is shown in figure 6.28.
6.14. SUMMARY
Megavoltage portal imaging, after years of being restricted to the use of film,
is now being approached seriously as a field for improvement. Electronic
portal imaging has become the state of the art, but has not yet quite reached
maturity, in that real-time portal imaging is not yet a routine reality in all centres.
Consequently, studies of the effectiveness of improved portal imaging are still
to be done. Several groups are beginning to use portal imagers to assess the
importance of field mispositioning with respect to the patient (Griffiths 1992,
Shalev 1992).
There is a variety of quite distinct technical instrumentation and it is difficult
to make quantitative comparisons with which the authors of each system would
agree. Swindell et a1 (1 991) have provided a very detailed study of the underlying
physics which brings out many of the important considerations, and this paper

Figure 6.27. The Royal Marsden Hospital megavoltage CT system. The

curved detector is attached to a Philips SL25 accelerator. This slots into the
triangular mounting plate which is firmly bolted to the rotating gantry of the
linac. (Courtesy of W Swindell, E Morton, D G Lewis and P Evans.)
Figure 6.28. A megavoltage CT scan taken on the system at the Royal

Marsden Hospital. The scan shows the external contour; the lungs and the
bones and soft-tissue detail can be resolved when viewing the image on a
monitor with a windowing faciliv. The two circular structures below the
patient are part of the patient support system. (Courtesy of W Swindell,
E Morton, D G Lewis and P Evans.)
should be appraised by the serious researcher. In this chapter we have discussed

some but not all of these aspects, whilst trying to present the operating features
of each system as simply and objectively as possible. This is a field which will
expand and indeed is only now beginning to attract serious commercial interest.
In summary, the need is for a detector which can operate in real-time, produce a

References 283
high resolution digital radiograph, has a high quantum efficiency, be of convenient

size, reject scattered radiation and possess a high signal-to-noise ratio so that the
low contrast inherent in portal radiology is challenged. At present, good detectors
exist, together with means to make accurate measurements. Attention should now
be tumed towards quantifying positional errors and using portal measurements to
compute the dose actually delivered to the patient.
REFERENCES
Amols H I and Lowinger T 1987 An inexpensive microcomputer based port film

image enhancement system Med. Phys. 14 483
Antonuk L E, Berry J, Boudry J, Huang W, Longo M J, Mody P, Morton E
3, Schick H, Seraji N, Yorkston J and Street R A 1992 Flat-panel imagers
for radiotherapy and diagnostic radiographic and fluoroscopic imaging Proc.
Megavoltage Imaging Meeting (Newport Beach, CA, 1992) (New York:
American Association of Physicists in Medicine)
Antonuk L E, Boudry J, Kim C W, Longo M, Morton E J and Yorkston J 1991a
Signal, noise, and readout considerations in the development of amorphous
silicon photodiode arrays for radiotherapy and diagnostic x-ray imaging Proc.
SPIE 1443 (Medical Imaging V: Image Physics) pp 108-1 19
Antonuk L E, Boudry J, Yorkston J, Wild C F, Longo M J and Street R A
1990a Radiation damage studies of amorphous silicon photodiode sensors for
applications in radiotherapy x-ray imaging Nucl. Instrum. Methods in Phys.
Res. A 299 143-146
Antonuk L E, Yorkston J, Boudry J, Longo M J, Jimenez J and Street R A 1990b
Development of hydrogenated amorphous silicon sensors for high energy
photon radiotherapy imaging IEEE Trans. Nucl. Sci. 37 (2) 165-170
Antonuk L E, Yorkston J, Boudry J, Longo M J and Street R A 1991b Large area
amorphous silicon photodiode arrays for radiotherapy and diagnostic imaging
Nucl. Instrum. Methods in Phys. Res. A 310 460464
Antonuk L E, Yorkston J, Morton E J, Boudry J, Kim C W and Longo M J 1991c
Flat-panel, self-scanning, solid state imagers for real-time radiotherapy and
digital fluoroscopy Med. Phys. 18 610
Baily N A, Horn R A and Kampp T D 1980 Fluoroscopic visualisation of
megavoltage therapeutic x ray beams Int. J . Rad. Oncol. B i d . Phys. 6 935-
939
Barrett H H ,Gordon S K and Hershel R S 1976 Statistical limitations in transaxial
tomography Comp. B i d . Med. 6 307-323
Barrett H H and Swindell W 1981 Radiological imaging: the theory of image
formation, detection and processing vol 1 pp 97-100,287-289
Benner S , Rosengren B, Wallman H and Netteland 0 1962 Television monitoring
of a 30 MV X-ray beam Phys. Med. Biol. 7 29-34

Bijhold J, Gilhuijs K G A, van Herk M and Meertens H 1991a Radiation field edge
detection in portal images Phys. Med. Biol. 36 1705-1710
Bijhold J, van Herk M, Vijlbrief R and Lebesque J V 1991b Fast evaluation of
patient set-up during radiotherapy by aligning features in portal and simulator
images Phys. Med. Biol. 36 1665-1679
Bijhold J, Lebesque J V, Hart A A M and Vijlbrief R E 1991c Maximising
setup accuracy using portal images as applied to a conformal boost technique
for prostatic cancer Proc. 1st biennial ESTRO meeting on physics in clinical
radiotherapy (Budapest, 1991) p 18
Biggs P J 1991 private communication (AAPM visiting fellow, May 21st 1991)
Biggs P J, Goitein M and Russell M D 1985 A diagnostic x-ray field verification
device for a 10 MV linear accelerator Int. J . Rad. Oncol. Biol Phys. 11635-643
Boag J W 1973 Xeroradiography Phys. Med. B i d . 18 3-37
Bova F J, Fitzgerald L T , Mauderli W M and Islam M K 1987 Real time
megavoltage imaging Med. Phys. 14 707
Boyer A L, Antonuk L, van Herk M, Meertens H, Munro P, Reinstein L E and
Wong J 1992 A review of electronic portal imaging devices (EPIDs) Med. Phys.
19 (1) 1-16
Brahme A, Lind B and Nafstadius P 1987 Radiotherapeutic computed tomography
with scanned photon beams Int. J . Rad. Oncol. Biol. Phys. 13 95-101
Bucciolini M, Pallotta S , Cionini L, Milano F and Renzi R 1991 Control of patient
setup during radiation treatments Proc. 1st biennial ESTRO meeting on physics
in clinical radiotherapy (Budapest, 1991) p 22
Crooks I and Fallone B G 1991 PC based selective histogram equalisation for
contrast enhancement of portal films Med. Phys. 18 618
Dance D R 1988 Diagnostic radiology with x-rays The physics of medical imaging
ed S Webb (Bristol: Adam Hilger) pp 2&73
Droege R T and Bjamgard B E 1979 Influence of metal screens on contrast in
megavoltage x-ray imaging Med. Phys. 6 487-493
Evans P M, Gildersleve J Q, Morton E J, Swindell W, Coles R, Ferraro M,
Rawlings C, Xiao Z R and Dyer J 1992a Image comparison techniques for
use with megavoltage imaging systems Brit. J . Radiol. 65 701-709
Evans P M, Gildersleve J Q, Rawlings C and Swindell W 1992b The
implementation of patient position correction using a megavoltage imaging
device on a linear accelerator Brit. J . Radiol. in press
Fingerhut A G and Fountinell P M 1974 Xeroradiography in a radiation therapy
department Cancer 34 78-82
Galkin B M, Wu R K and Suntharalingam N 1978 Improved technique for
obtaining teletherapy portal radiographs with high-energy photons Radiology
127828-830
Gildersleve J 1992 Clinical aspects of megavoltage imaging MD thesis University
of London
Gildersleve J, Deamaley D P, Evans P M, Law M, Rawlings C and Swindell W
1992a A randomised trial of patient repositioning during pelvic radiotherapy

References 285
Proc. BIR Con5 on Megavoltage Portal Imaging (London, 1992) (London:

BIR)
Gildersleve J, Evans P M, Deamaley D P, Rawlings C and Swindell W
1992b Reproducibility of patient positioning, as assessed by an Integrated
Megavoltage Imaging system Proc. BIR Con5 on Megavoltage Portal Imaging
(London, 1992) (London: BIR)
Gildersleve J, Swindell W, Evans P, Morton E, Rawlings C and Deamaley D 1992c
Verification of patient positioning during radiotherapy using an integrated
megavoltage imaging system Proc. ART91 (Munich) (abstract book p 135)
Advanced Radiation Therapy: Tumour Response Monitoring and Treatment
Planning ed A Breit (Berlin: Springer) pp 693-695
Graham M L, Cheng A Y, Geer L Y, Binns W R, Vannier M W and Wong J W 1991
A method to analyze 2-dimensional daily radiotherapy portal images from an
on-line fiber-optic imaging system Int. J . Rad. Oncol. Phys. 20 613-619
Griffiths S 1992 Quality assurance-what is possible? Proc. 50th Ann. Congress
of the British Institute of Radiology (Birmingham, 1992) (London: SIR) p 29
Grimm D F, Wilson C R and Gillin M T 1987 Digital processing of radiotherapy
port films Med. Phys. 14 483
Halmshaw R 1966 Physics of industrial radiography (London: Heywood) p 488
Hammoudah M M and Henschke U K 1977 Supervoltage beam films Int. J . Rad.
Hendee W R 1978 Computed tomography in radiation therapy treatment planning
Int. J . Rad. Oncol. B i d . Phys. 4 539-540
van Herk M 1991 Physical aspects of a liquid-filled ionization chamber with
pulsed polarizing voltage Med. Phys. 18 (4) 692-702
- 1992 An electronic portal imaging device: physics, development and
application PhD thesis University of Amsterdam
van Herk M and Meertens H 1987 A digital imaging system for portal verification
The use of computers in radiation therapy ed I A D Bruinvis et a1 (Roc.9th
ICCRT) pp 371-373
-1988 A matrix ionisation chamber imaging device for on-line patient setup
verification during radiotherapy Radiother. Oncol. 11369-378
Holloway A F 1958 A localising device for a rotating cobalt therapy unit Brit. J .
Radiol. 3 1 227
Hoogervorst B R, van Herk M and Mijnheer B J 1991 Development of a two-
dimensional array dosimeter Proc. 1st biennial ESTRO meeting on physics in
Jevbratt L, Lagergren C and Sarby B 1971 Direct beam control in radiotherapy
with high energy photons Acta Radiol. Ther. Phys. B i d . 10 433-442
Jones S M and Boyer A L 1991 Investigation of an FFT-based correlation
technique for verification of radiation treatment setup Med. Phys. 18 (6) 1116-
1125
Kiillman P, Lind B, Iacobeus C and Brahme A 1989 A new detector for
radiotherapy computed tomography verification and transit dosimetry Proc.

17th Int. Congress on Radiology (Paris, 1989) p 83

Kanamori T, Kamata S and Ito S 1989 Cross-sectional imaging of large and dense
materials by high energy x-ray CT using linear accelerator J. Nucl. Sci. Tech.
26 (9) 826-832
Keyes W I 1975 The fan-beam gamma camera Phys. Med. Biol. 20 489-493
Lam K S , Partowmah M and Lam W C 1986 An on-line electronic portal imaging
system for extemal beam radiotherapy Brit. J. Radiol. 59 1007-1013
Leong J 1984 A digital image processing system for high energy x-ray portal
images Phys. Med. Biol. 29 1527-1535
Leong J 1986 Use of digital fluoroscopy as an on-line verification device in
radiation therapy Phys. Med. Biol. 31 985-992
Leszczynski K, Miskiewicz K, Cosby S , Shalev S , Reinstein L and Meek A
1988 On-line megavoltage imaging 2: Image processing Electronic Imaging
88; International electronic imaging exposition and conference (World Trade
Centre, Boston, 1988) (Waltham: Institute for Graphic Communications) pp
1080-1 083
Leszczynski K, Shalev S and Cosby N S 1990 An adaptive technique for digital
noise suppression in on-line portal imaging Phys. Med. Biol. 35 429-439
Lewis D G, Morton E J and Swindell W 1988 Precision in radiotherapy: a linear
accelerator based CT system Megavoltage Radiotherapy 1937-1 987 Brit. J .
Radiol. (Suppl. 22) p 24
Lewis D G and Swindell W 1987 A megavoltage CT scanner for radiotherapy
verification The use of computers in radiation therapy ed I A D Bruinvis et a1
(P~oc.9th ICCRT) pp 339-340
Lewis D G, Swindell W, Morton E, Evans P and Xiao Z R 1992 A megavoltage
CT scanner for radiotherapy verification Phys. Med. Biol. 37 1985-1999
Lutz W and Bjamgard B 1985 A test object for evaluation of portal films Int. J.
Rad. Oncol. Bioi. Phys. 11 631-634
Meertens H 1985 Digital processing of high-energy photon beam images Med.
Phys. 12 111-113
Meertens H 1989 On line acquisition and analysis of portal images PhD thesis
University of Amsterdam
Meertens H, Bijhold J and Strackee J 1990a A method for the measurement of
field placement errors in digital portal images Phys. Med. Biol. 35 299-323
Meertens H, van Herk M, Bijhold J and Bartelink H 1990b First clinical experience
with a newly developed electronic portal imaging device Int. J . Rad. Oncol.
Biol. Phys. 18 1173-1181
Meertens H, van Herk M and Weeda J 1985 A liquid ionisation detector for digital
radiography of therapeutic megavoltage photon beams Phys. Med. Biol. 30
3 13-32 1
Morton E J 1988 A digital system for the production of radiotherapy verification
images PhD thesis University of London
Morton E J, Evans P M, Ferraro M, Young E F and Swindell W 1991a
Development of video frame store and distortion correction facilities for an

References 287
extemal-beam radiotherapy treatment simulator Brit. J. Radiol. 64 747-750

Morton E J, Lewis D G and Swindell W 1988 A method for the assessment of
radiotherapy treatment precision Megavoltage Radiotherapy 1937-1 987 Brit.
J . Radiol. (Suppl. 22) p 25
Morton E J and Swindell W 1988 A digital system for the production of
radiotherapy verification images The use of computers in radiation therapy ed
I A D Bruinvis et a1 (Proc. 9th ICCRT) pp 375-377
Morton E J, Swindell W, Lewis D G and Evans P M 1991b A linear array
scintillation crystal-photodiode detector for megavoltage imaging Med. Phys.
18 (4) 68 1-69 1
Munro P M, Rawlinson J A and Fenster A 1990a A digital fluoroscopic imaging
device for radiotherapy localisation Int. J. Rad. Oncol. Biol. Phys. 18 641-649
-1990b Therapy imaging: a signal-to-noise analysis of a fluoroscopic imaging
system for radiotherapy localisation Med. Phys. 17 (5) 763-772
Nakagawa K, Aoki Y, Akanuma A, Onogi Y, Karasawa K, Terahara A, Hasezawa
K and Sasaki Y 1991 Development of a megavoltage CT scanner using linear
accelerator treatment beam J. Japan. Soc. Therap. Radiol. Oncol. 3 265-276
Reinstein L E and Orton C G 1978 Contrast enhancement of high energy
radiotherapy films Med. Phys. 5 332
-1979 Contrast enhancement of high energy radiotherapy films Brit. J. Radiol.
52 880-887
Rose A 1948 The sensitivity performance of the human eye on an absolute scale
J . Opt. Soc. Am. 38 196-208
-1973 V i s i o A u m a n and electronic (New York: Plenum)
Rudin S , Bednarek D R and Wong R 1991 Accurate characterization of image
intensifier distortion Med. Phys. 18 (6) 1145-1 151
Scheck R J, Wendt Th and Panzer M 1992 Digital portal radiography in radiation
therapy Proc. BIR Conf. on Megavoltage Portal Imaging (London, 1992)
(London: BIR)
Shalev S 1991 The physical limitations of DQE in portal imaging systems Med.
Phys. 18 616
-1992 Report on 2nd Annual Int. Workshop on Electronic Portal Imaging
(Newport Beach, 1992) (private communication)
Shalev S , Lee T, Leszczynski K, Cosby S , Chu T and Reinstein L 1989 Video
techniques for on-line portal imaging Comp.Med. Im. Graph. 13 217-226
Shalev S, Leszczynski K, Reinstein L and Meek A 1988 On-line megavoltage
imaging 1: system design Electronic Imaging 88; International electronic
imaging exposition and conference (World Trade Centre, Boston, 1988)
(Waltham: Institute for Graphic Communications) pp 1074-1079
Shalev S , Ryder S, Cosby S and Miskiewicz K 1991 Clinical experience with on-
line portal imaging Proc. ART91 (Munich) p 138
Shorvon L M, Robson N L K and Day M J 1966 A new technique for treatment
planning and field verification Clin. Radiol. 17 139-140
Simpson R G, Chen C T, Grubbs E A and Swindell W 1982 A 4MV CT scanner

for radiation therapy: The prototype system Med. Phys. 9 (4) 574-579
Smith V 1987 Routines for enhancement of radiation therapy images The use of
computers in radiation therapy ed I A D Bruinvis et a1 (Proc. 9th ICCRT) pp
3740
Smith A R 1990 Evaluation of new radiation oncology technology Int. J. Rad.
Oncol. B i d . Phys. 18 701-703
Sonoda M, Takano M, Miyahara J and Kato H 1983 Computed radiography
utilising scanning laser stimulated luminescence Radiology 148 833-838
Sontag M and Purdy J A 1991 (Writing Chairs) State-of-the-art extemal photon
beam radiation treatment planning Int. J . Rad. Oncol. Bid. Phys. 21 9-23
Swindell W 1991 The lens coupling efficiency in megavoltage imaging Med. Phys.
18 1152-1153
- 1992a Transit dosimetry Proc. BIR Conference on Megavoltage Portal
Imaging (London, 1992) (London: BIR)
- 1992b Imaging and portal verification Review paper at the 50th Ann.
Congress of the British Institute of Radiology (Birmingham, 1992) (London:
BIR) p 29
Swindell W and Gildersleve J 1991 Megavoltage imaging in radiotherapy Rad.
Magazine (Sept 1991) 17 (196) 18-20
Swindell W, Lewis D G and Morton E J 1988Radiological imaging methods based
on a linear accelerator: implementation and application Brit. J. Radiol. 61 7 17-
718
Swindell W, Morton E J, Evans P M and Lewis D G 1991 The design of
megavoltage projection imaging systems: some theoretical aspects Med. Phys.
18 855-866
Swindell W, Simpson R G and Oleson J R 1983 Computed tomography with a
linear accelerator with radiotherapy applications Med. Phys. 10 (4) 416-420
Taborsky S C, Lam W C, Stemer R E and Skarda G M 1982 Digital imaging for
radiation therapy verification Optical Eng. 21 888-893
Varian 1992 Varian Portalvision Varian Associates product literature
Visser A G V, Huizenga H, Althof V G M and Swanenburg B N 1990 Performance
of a prototype fluoroscopic radiotherapy imaging system Int. J . Rad. Oncol.
B i d . Phys. 18 43-50
Wade J P and Nicholas D 1991 Clinical application of an electronic imaging device
for assisting patient set-up in radiotherapy Brit. J . Radiol. 64 59&602
Webb S 1990 Non-standard CT scanners: their role in radiotherapy Int. J . Rad.
Oncol. B i d . Phys. 19 1589-1607
Webb S, Binnie D M, Flower M A and Ott R J 1992 Montecarlo modelling of the
performance of a rotating slit-collimator for improved planar gamma-camera
performance Phys. Med. B i d . 37 1095-1108
Webb S and Johnson C V 1990 The perception of detail in radioisotope images
with Gaussian noise Phys. Med. Biol. 35 1145-1151
Wilenzick R M, Memtt C R B and Balter S 1987 Megavoltage portal films using
computed radiographic imaging with photostimulable phosphors Med. Phys.

References 289
14 389-392
Wolfe J N, Kalisher L and Considine B 1973 Cobalt 60 treatment field verification
by xeroradiography Am. J . Ront. 118 916-918
Wong J W, Binns W R, Cheng A Y, Epstein J W and Klarmann J 1991
Development of a high resolution fiber-optic radiotherapy imaging system
Med. Phys. 18 609
Wong J W, Binns W R, Cheng A Y, Geer L Y, Epstein J W, Klarmann J and Purdy
J A 1990a On-line radiotherapy imaging with an array of fibre-optic image
reducers Znt. J. Rad. Oncol. Biol. Phys. 18 1477-1484
Wong J W, Slessinger E D, Hermes R E, Offitt C J, Roy T and Vannier M W 1990b
Portal dose images I: Quantitative treatment plan verificationZnt.J.Rad. Oncol.
Biol.Phys. 18 1455-1463
Wowk B, Shalev S and Leszczynski K W 1991 Optimisation of metal/phosphor
screens for on-line portal imaging Med. Phys. 18 616
Ying X, Geer L Y and Wong J W 1990 Portal dose images 11: Patient dose
estimation Int. J.Rad. Oncol. Biol.Phys. 18 1465-1475

CHAPTER 7
TREATMENT MACHINE FEATURES

FOR CONFORMAL THERAPY
7.1. THE EARLIEST TREATMENT MACHINE FOR CONFORMAL

THERAPY WITH A 13’CS SOURCE
In this chapter we briefly review some of the machine technology for conformal
therapy and gather together some of the ‘accessories’ found on such machines. By
way of introduction, probably the first radiotherapy treatment machine to make use
of a high-activity 137Cssource was designed to investigate the optimal geometrical
movements for ‘concentrating’ radiation in a small volume of the patient (Brucer
1956). This early machine for conformal therapy, constructed for the Oak Ridge
Institute for Nuclear Studies, Tennessee, by W F and John Barnes CO,Rockford,
Illinois, was loaded with 1540 Ci of the reactor fission waste product 137Cs.The
treatment machine was never intended as a routine clinical tool, but designed to
investigate the relative merits of conical, pendulum and rotational therapy (see
appendix 2A). The source could make any movement over the surface of a sphere
whose radius was variable between 50 and 90 cm, with the beam forever directed
at a central fixed point in space. Meanwhile the patient couch, a highly modified
surgical table, could also execute a variety of translations and rotations. The aim
was to develop a test bed, whereby a variety of patterns of irradiation were studied
independent of the practicality of delivery in the clinic. Brucer (1956) wrote ‘The
literature of radiation therapy is full of therapeutic decisions that actually were
forced by the limitations of a piece of machinery. The purpose of this excedingly
complex device is thus to study therapeutic procedures unhampered by the bias of
mechanical restrictions.’
Movements were controlled by analogue computer with motor-driven cams
controlling the beam trajectory. Movements of a wedge filter could be coordinated
with the movement of the source to accommodate the change in depth of the
tumour with changing beam trajectory. The beam could be switched on and off for
parts of the orbit via further mechanical cams. In view of the potential danger of
collisions, touch-sensitive motor cut-outs and beam shielding was provided. The
machine maintained an accuracy, which would today be regarded as rather poor,
290
Tracking units 291
motion
Table
t ranslatlon
t-
Source trqec t o r y
relatlve to patient
Figure 7.1. An example of conformal therapy achieved by dynamic means.

The linac gantry rotates and the couch translates. The jaws of the accelerator
adopt different widths depending on the view of the target. The treatment
shown is for retroperitoneal nodes and pelvic sidewalls. (From Levene et a1
(1978).)
of less than 1 cm, largely set by the design of the cams.
7.2. TRACKING UNITS
A high-dose volume can be made to conform with an irregularly shaped 3D

target by dividing the irradiation into a large number of contiguous or slightly
overlapping elemental fields. The isocentre is made to follow a complicated locus
through the patient. This is the principle of ‘tracking therapy units’. In such
units the couch can move laterally and vertically, and of course the gantry can
rotate. These movements, with the beam on, can be continuous. The method
is also sometimes called ‘dynamic therapy’. The subject has a long history. At
some centres tracking units have been based on the linear accelerator (Chung-Bin
and Wachtor 1979, B j h g a r d et a/ 1977, Levene et a1 1978, Chin et a1 1981,
Saunders and Chin 1986). Figure 7.1 shows how the high-dose volume could
be conformed to an irregularly shaped volume by rotating the treatment head of a
linear accelerator with the jaws adopting a varying width and the couch translating.
The name of the Royal Free Hospital (RFH) in London is always associated in
the UK with the ‘tracking cobalt unit’. The first developments were made in the
1950’s (Jennings 1985) and later versions of the machine are still being developed
(Brace 1985, Davy 1987).
With a cobalt machine, the dose is proportional to the time spent irradiating
a point, or altematively inversely proportional to the tracking speed at which
the point passes the beam. Thus exposure-time profiles (ETP) can specify the
irradiation, leading to absorbed-dose profiles (ADP). For example, several non-
uniform ETPs designed to shield organs at risk might be combined to give a
uniform ADP in some target volume (Davy 1985). Tracking machines may be

292 Treatment machine features for conformal therapy
used in many different modes, depending on which components are actually in

motion.
Tracking units have perhaps not become as popular as hoped by their pioneers.
A number of reasons may be advanced. Certainly the treatment planning is
complex, generally unique to each unit, and thus somewhat non-mainstream. For
this reason (and also because excellent descriptions can be found elsewhere (Davy
1985)) this will not be discussed in detail here. Equally, verification, important for
all radiotherapy, is particularly important for tracking therapy and is complicated.
Shentall and Brace (1987) have described how the treatment planning for the
RFH tracking unit is performed for spinal axis irradiation. The planning process
optimizes the velocity of the longitudinal couch speed and includes an end-of-track
feature to avoid the large penumbra (equivalent to the field size) which would be
associated with having a constant couch speed and fixed field size at the beginning
and end of the track. This end-of-track method is, to start with, the minimum field
length in the direction of tracking and increases the field size at twice the speed
of the couch movement (and reverses the process at the cessation of irradiation).
At the start, this means the trailing edge of the field stays stationary relative to the
patient (small penumbra). The leading edge creates a wide penumbra, which is
cancelled exactly by the subsequent movement of the full-width field.
Blake et a1 (1987) have shown how combining dynamic collimator motion with
a static patient is another way to produce the same effect as moving the patient
relative to a fixed small field. Computer-controlled dynamic collimation is a
feature of the Philips SL-25 linear accelerator. One pair of collimator jaws can
cross the centreline by 12.5 cm. Imagine they are set to a small separation and
swept at constant velocity. The resulting dose profile is shown in figure 7.2. At
either end, the dose ramps to the maximum dose. This large penumbra is equal to
the collimator aperture and for practical reasons the end-of-track technique for the
RFH cobalt unit cannot be used. The penumbra can be reduced to no more than
obtained with a large single static field by appending extra field segments with just
one jaw moving.
Richter et a1 (1987) achieve dynamic conformal therapy by a combination
of computer control of the linac jaws, collimator and gantry rotation and four
movements of the patient support system (couch translation-three orthogonal
movements-and isocentric rotation.) The machine is a Philips SL 75/20.
Schlegel et a1 (1987) control the couch movements computationally and they
also control the shape of the field by dynamically adjusting a multileaf collimator.
The equipment is a Siemens Mevatron-77 linac. The planning system treats
dynamic therapy as a series of multiple sub-beams.
7.3. A TRACKING LINAC WITH MULTILEAF COLLIMATOR AND CT

COMBINATION
Nakagawa et a1 ( I 987) have engineered to add computer-controlled longitudinal

The universal wedge for the linear accelerator 293
Distance Along Swept Length
Figure 7.2. Showing the dose profile resulting j?om a pair of scanned
asymmetric jaws. The two approximately straight lines show the position of
the twojaws as a function of accumulated dose. The lower line shows the dose
delivered to each point as a function of distance. (From Blake et a1 (1987).)
couch movement to a linear accelerator facility. The distinguishing feature of the

development is that there is a CT scanner, in the same room as the linac, sharing
the same treatment couch. The scanner and linac face each other (figure 7.3). The
patient is scanned in the treatment position and the target volume is determined
in each CT section. The linear accelerator (Mitsubishi ML-20M) is equipped with
a multileaf collimator (Mitsubishi Multi- 16). With the patient couch re-oriented
and acting as the treatment couch, the width of two or three open pairs of leaves
is dynamically adjusted to the projected width of the target as each slice passes
through the beam, with the couch translating longitudinally. The gantry angle can
also be dynamically controlled by computer.
7.4. THE UNIVERSAL WEDGE FOR THE LINEAR ACCELERATOR
Wedge filters are traditionally used in radiotherapy to assist shaping the dose
distribution. For many accelerators a range of removable wedges are available.
The wedge angle 0, is defined (ICRU 1976) as the angle between the normal to
the central axis and the line tangential to the isodose curve at a depth of 10 cm.
It is more convenient, and possibly avoids errors, if the accelerator is fitted with
a universal wedge giving a fixed wedge angle 0,. For example, the Philips SLP5
accelerator has a universal wedge of nominally 0, = 60". The universal wedge
avoids the need to enter a treatment room to set a removal wedge in place, means
only one wedge field needs to be measured at commissioning and may be more
accurately engineered. By treating a patient for part of each fraction with the

Figure 7.3. The arrangement in Tokyo with a linear accelerator (equipped

with a multileaf collimator and computer-controlled moving couch} in the
same room as a CT scanner for target localization. (From Nakagawa et a1
( I 987}.}
universal-wedged field and for part of each fraction with an open field, a wedged
distribution with any wedge angle can be delivered.
Let D , be the depth-dose distribution with such a wedge in place and DObe the
depth-dose distribution with no wedge. A dose distribution D with a wedge angle
e,, less than Ow, may be constructed by combining a fraction B of the (universal)
wedged field with a fraction (1 - B ) of the open unwedged field; i.e.
D = (1 - B)Do + BDW. (7.1)
Petti and Siddon (1985) showed that the fraction B should be determined from
B = f / [(tanO,/tanO,) +f - 11 (7.2)
where
aDo/az
f =- (7.3)
aDw/az
is the ratio of the slopes of the central axis depth-dose curves for the open and
universal-wedged fields (at 10 cm depth). Equation (7.2) is analytically correct;
Petti and Siddon (1985) give the derivation.
They showed, from measured data, that for fields smaller than 15 x 15 cm’, f
was almost unity, but for large fields such as 20 x 20 cm*, f departed from unity
by as much as 20%,indicating the use of the full equation (7.2). Under conditions
where f = 1, equation (7.2) reduces to
B = tanO,/tanO, (7.4)

The dynamic wedge for the linear accelerator 295
a result proposed by Philips Medical Systems Division (1983) and discussed by

Zwicker et a1 (1985). Approximate equation (7.4) predicts 0, to within 3" for fields
up to 20 x 20 cm2 and even better for smaller fields. A formula due to Tatcher
(1970)
B = e,p, (7.5)
is inadequate when the universal wedge angle is (as in Petti and Siddon's study)
60", but accurately predicts 0, to about a degree if 0, is less than or equal to
30". Mansfield et a1 (1974), however, appeared to verify Tatcher's equation for
6oCoradiation and a 60" universal wedge. In practice, the fraction B is usually
determined so that 0, matches the wedge angle for some particular removable
wedge, allowing the planning experience gained with removable wedges to carry
over to the universal wedge. However, the two situations may not be entirely
identical because of changes in the scattered dose (McParland 1990).
Both McParland (1991) and Petti and Siddon (1985) showed the effective
wedge angle 0, itself increased with field size by about 0.5"-0.75" cm-' for a
Philips/ MEL SL75/5 accelerator equipped with a universal wedge of nominal
angle 60". McParland (1991) quoted the increase was from 49" at 5 x 5 cm2 to
60" at 20 x 20 cm2. This wedge was 81 cm from the isocentre and made of 96%
lead and 4% antimony alloy.
McParland (1991) studied the effect of field size and also angulation of the
universal-wedged fields on the dose to the build-up region. The former was a
small effect, but the latter was noteworthy.
A problem arises in treatment planning in that it may not be at all clear ahead
of time what would be the best wedge angle to choose for a particular set of
circumstances. The problem is worse if a number of fields are to be combined to
achieve a conformal plan. The problem can be specified: for N portals, with each
portal comprising two fields-one open and one wedged with the fixed wedge
angle-what are the 2N weights to be attached to each field? Cheng and Chin
(1987) have shown that, by specifying the dose to a predefined set of points in the
3D volume and defining a suitable cost function, the problem can be cast as an
optimization problem and so solved.
7.5. THE DYNAMIC WEDGE FOR THE LINEAR ACCELERATOR
Another way to produce a symmetric field with isodose contours angled relative to
the axis of symmetry, without using either a fixed extemal wedge or a combination
of a fixed wedge and an open field (the universal wedge), is the so-called 'dynamic
wedge'. The dynamic wedge is the name given to the technique whereby a series
of asymmetric fields of different sizes are combined. One jaw of the accelerator
collimator remains fixed for all these sub-fields; the other moves to a series of
locations varying from the full width of the required wedged field down to close
to the first collimator jaw position. The static asymmetric sub-fields are delivered

-10 -5 0 5 10
DISTANCE FROM CENTRAL AXISkml
Figure 7.4. Isodose distributions for a 20 cm wide mixed-wedge field (45'

over half the field, 0" over the other half). The wedge angle is defined at a
depth of I O cm. Distributions of this kind can be obtained by using a dynamic
wedge. (From Leavitt et a1 (19901.)
with a different number of monitor units per sub-field; the distribution of these
values determines the wedge angle.
Leavitt et a1 (1990) describe implementing this technique for a Varian Clinac
2100C accelerator. The full-field segment is delivered first, followed by each
smaller field in tum. In principle, the dose segments can be delivered for changes
in collimator position as small as 1 mm, but Leavitt et a1 (1990) found increments
of 5 mm were acceptable for routine use, giving the same dose distribution as
obtained with smaller increments.
The advantages claimed for the dynamic wedge are:
e The prescribed isodose line at the wedge angle extends over a greater fraction
of the field than can be achieved using a standard fixed wedge.
e The wedge angle can be specified arbitrarily and at any depth.
e So-called 'customized wedge shapes' can be delivered with the wedge angle
varying across the field, even being zero for part of the field, as shown in figure
7.4.
Set against these are the potential disadvantage of increased time per field
because one jaw has to move and have its position verified.
The number of monitor units for each asymmetric field is computed as follows:
The computer is told the desired wedge angle and central axis depth at which this
is defined. Reference points are distributed along the line corresponding to this
idealized wedge isodose line, there being as many reference points as sub-fields,
i.e. the first reference point receives only primary dose from the full-width field;
the second receives primary dose from the first and second field and so on until the
last reference point receives dose from only the smallest asymmetric field. Hence
the number of monitor units for the first field must give the entire dose required
by the first reference point (actually reduced by 2% because some scatter from
other points will contribute). The number of monitor units for the second field has

Linear accelerators with independent collimators 297
to provide the difference between what the second reference point requires and
what it has already received from the first sub-field.. . and so on. The weights
so determined may be fine tuned by an optimization technique and it turns out
that the monitor units per setting does not vary smoothly between fields, as would
be the case if the jaw moved at constant velocity. The monitor unit weights for
customized fields can be similarly found.
The set of asymmetric fields is delivered automatically under computer control.
A dynamic wedge is a standard feature of Varian linear accelerators, in which
it may travel continously at constant velocity if required.
7.6. WEDGES WITH THE MULTILEAF COLLIMATOR ON A LINEAR

ACCELERATOR
When a multileaf collimator is attached to a linear accelerator it is sometimes

necessary to locate the field wedge on the patient side of the blocking tray. This
minimizes the distance from the source to the blocking tray and can maintain this
distance the same as without the MLC. Ochran et a1 (1992) studied this for a Varian
linear accelerator equipped with a multileaf collimator. Their study came to the
following conclusions:
1. The shape of the wedged isodose distributions exhibited no essential
differences from when the wedges were mounted above the blocking tray closer
to the source.
2. The average difference between the nominal wedge angle and the observed
wedge angle was 2" at 18 MV and 5.5" at 6 MV for the four wedges of nominal
angles 15", 30", 45" and 60". The measured angle was made between the isodose
line and the normal to the central axis at 10 cm depth.
3. The wedge factor (defined as the output for the wedged field divided by the
output for the open field) was measured at d, and at 5 and 10 cm depth. The
wedge factor vaned with beam energy, field size and wedge angle, as expected,
but had little depth dependence.
4. The surface dose with the wedge below the blocking tray was larger than
when it was above. The surface dose increased with decreasing SSD and as the
field size increased, consistent with the theory that this was due to increasing low-
energy scattered radiation.
7.7. LINEAR ACCELERATORS WITH INDEPENDENT COLLIMATORS
The modem multileaf collimator is sufficiently important to conformal therapy

to have been discussed at length earlier in a separate chapter. Traditionally,
however, linear accelerators have two sets of jaws, orthogonal to each other which
can collimate square or rectangular fields symmetrical about the central beam
axis. Recently some manufacturers have introduced independent collimators.

F
Figure 7.5. Showing the nomenclature for beams. F is the source, SSD is
the source-to-skin distance, d is the depth of a measurement point B on axis.
Off-axispoints may be specified by either the distance X or the angle 4. 40
would refer to the angle made by the central axis of an asymmetrical field.
(From Thomas and Thomas (1990).)
The jaws can be moved separately, cross the central axis and thus may define
rectangular asymmetric fields. A number of studies have been made of the
dosimetry, including Chui and Mohan (1986), Khan et a1 (1986) and Loshek and
Keller (1988). The latter showed that an asymmetric field could be treated as a
blocked symmetric field.
Alternatively, the dose at a point in such circumstances can be computed
directly. Thomas and Thomas (1990) have shown this may be expressed as
where d (S, d , 4, 40)is the dose per monitor unit for a field of area S at depth d ,
at position labelled by angle 4, asymmetrically centred at angle $0 (figure 7.5).
DD,, (S, d ) is the on-central-axis depth dose (dose at depth d divided by dose at
build-up depth dmm) for the same area field at the same depth, but symmetrically
placed about the central ray of the machine. 0 Fsp (S) is the output factor (dose
per monitor unit at d,,,) on the central axis for the symmetrically placed field.
OCR,,, (4, d ) is the off-centre ratio (ratio of the dose at a point off the central
ray to the dose at the same depth on the central ray) at angle 4 from the central
ray, at depth d for a symmetrically placed field. P 0 C R (4, d ) is the primary off-
centre ratio at angle # and depth d (which is the ratio of the dose due to primary
radiation at a point off-axis to the primary radiation at the corresponding depth on-
axis). P OC R (+, d ) is given in terms of the in-air POCR at the SSD, P OC R (+, 0)
by
POCR ( 4 , 4 = P O C R ( + , O ) e x p ( - ~ ( 4 ) d / c o s 4 ) / e x p ( - ~ o d ) (7.7)
and we note that F is a function of angle 4, indicating the hardening of the beam
off-axis. 10is the coefficient along the axis of the machine.

Two-dimensional tissue compensators 299
The depth-dose, output factor and off-axis ratio are measured routinely for
symmetrical fields. The above analysis shows that the only other data needed
to compute the dose distribution for asymmetrical fields is the measurement of the
in-air POCR and the angular dependence of the beam hardening.
Thomas and Thomas (1990) measured the necessary data and showed that these
two equations predicted the dose accurately when compared with measurements
for a Brown-Boveri CH8 8 MV linac. The beam hardening was expressed by
p (4) = 0.037 + 0.0204 (7.8)

with p in cm-’ and 4 in radians.
Independently movable jaws crossing the mid-line have also be used to generate
a wedged field (Levene et a f 1978) similar to that produced by the ‘dynamic
wedge’.
7.8. TWO-DIMENSIONAL TISSUE COMPENSATORS
A two-dimensional tissue compensator is a material placed in the path of a

beam, usually in the ‘blocking tray’ whose thickness varies across its area in a
way designed to overcome the effects of obliquity of the beam, the non-flat patient
contour and internal tissue inhomogeneities. A wedge is a particularly simple
form of such a compensator and correspondingly achieves a simple distribution
of dose. In contrast, a tissue compensator can be designed to give a uniform
dose on a specified plane within the patient. This has related objectives to
conformal therapy. For example, combination of several beams, each with its
own compensator, would be a desirable objective (see later this section).
For the moment consider a single beam and the objective of obtaining a uniform
dose to a specified plane, which, in general, need not be normal to the beam
axis. Mageras et al (1991) have described a method of computing the shape of
the compensator.
The compensator is specified by its transmission T ( x , y). The thickness of the
compensator is then
( x , Y> = -In T ( x , Y) C
IL (7.9)
where p is the linear attenuation coefficient of the compensator material. For the
purposes of computation, the transmission matrix T is imagined to be centred at
the isocentre and perpendicular to the z central axis of the beam. The required
plane of constant dose is shown shaded in figure 7.6. A point in this plane is
labelled xp,yp and a ray from the source to this point intersects the transmission
matrix at x , y. The process of computing T ( x , y) is iterative:
0 Compute the dose to xp,yp from
DXP& - & T A R ( d , wp x Hp) O C R ( d , xp, yp, wp,Hp) (s,~/s>~ C

(7.10)

constant aosc
Figure 7.6. The geometry for computing the thickness of a 2 0 compensator

needed to generate a uniform dose distribution over the dark-shaded plane.
The compensator is shown in the blocking tray. The aperture is defined by the
beam’s-eye-view of the target and the transmission matrix is imagined to be
at the isocentrefor the purposes of computation (see text). (From Mageras el
a1 (19911.)
where
D,f is the reference dose at distance sxf from the source,
d is the water-equivalent depth of xp, yp from the source (computed from 3D
CT data),
s is the distance of the point from the source,
T A R ( d , W, x H,) is the tissue-air ratio at depth d,
W,, Hp is the field size at xpry,,,
0 C R is the off-centre ratio, and
C is a correction factor which accounts for the change in the fluence (and hence
dose) due to the compensator (and blocks etc).
0 C is a ratio of two doses at the point x , y , d, in a flat-surfaced uniform phantom,
one with the fluence I$compmodified by the compensator transmission and the
other with the fluence I$openbeing the uniform fluence within the beam’s-eye-
view of the plane; i.e.

Two-dimensional tissue compensators 301
The function k is the pencil-beam dose distribution.

0 For the first iteration the fluence dcomp
is assumed to be unmodified (=dopen)
and
C = 1. Then the first estimate of the transmission is
T“’ ( x , y) = D::;,,/D‘” ( x , y) (7.12)
where the superscript ( I ) has been added to show this is the first estimate. D:;hst
is chosen so the maximum transmission is unity.
0 Now the process is repeated, except that this time the fluence dcomp in the
numerator of C is obtained from the matrix of transmissions T . C now departs
from unity and a new dose D(2)is computed from which the transmissions are
adjusted to ( x , y dependence not explicitly shown)
The calculations are of course made for each point in the plane where it is
required to have uniform dose. After several iterations, (and Mageras et al
(199 1) suggest only some two or three are required in practice), the transmission
matrix becomes:
) T‘k-l)D(k)
~ ( k= const/D(k)’ (7.14)
Because of the use of the pencil-beam kemel, this iterative process has
accounted for more than just the effect of changed primary fluence on the dose and
included the three-dimensional transport of radiation within the patient. Mageras
et a1 (1991) used this technique to compute the thicknesses of compensator and to
drive an automatic milling machine to fabricate the compensator. Film dosimetry
then verified that the method could accurately flatten the dose in a plane to within
a few percent. Renner et al(1989) tackle this problem similarly, but do not use
the convolution method of including energy transport.
Ideally one wishes to combine beams, each equipped with its own compensator,
such that the resulting combination of doses generates a distribution which is
optimal. For conformal radiotherapy this means creating as uniform a target dose
as possible whilst minimizing the dose to other healthy tissue. Djordjevich et
a1 (1990) formulated the solution to this problem. Each beam is divided into
a number of small elementary beams; the problem is to compute the weight of
these elementary beams. Let the weight of the j t h elementary beam be W j and tij
indicate the contribution to the ith target point from the jth elementary beam. The
total dose to the ith target point from N elementary beams is then
AI
Tdi = c t i j w j , i = l , 2 ... mi (7.15)

j=1
for mi target points. Similarly, let ui, indicate the contribution to the ith point in
an organ at risk from the jth elementary beam. The total dose to the ith point in

an organ at risk from N elementary beams is then
N
'di = c u i j w j , i = 1 , 2 ... mu (7.16)
j=l
for m u organ at risk points. Suppose the prescribed dose to the ith target point is
ri, then the problem becomes how to minimize
x 2 = (TD - R)' (TD - R ) + AY' ("D) (7.17)
where ' signifies transpose, bold capitals indicate vectors of the corresponding
lower case components and Y is a vector which weights the importance of each
healthy (organ at risk) point. A is a multiplier which may be varied to control the
relative importance of obtaining a homogeneous dose distribution in the target,
compared with the importance of minimizing the dose to organs at risk. A further
constraint to control the range of weights was applied; for example, weights
clearly must be positive. Djordjevich er a1 (1990) show that minimizing x2 is
a well known problem in quadratic programming, and they obtain solutions for
model planning problems.
Although formalism of this kind has been discussed in terms of designing a set
of two-dimensional tissue compensators, the same arguments apply to designing
the weights of elemental sub-beams within an irregularly shaped field determined
by a multileaf collimator. The solution of the inverse problem 'given the required
dose constraints, find the intensity distribution across the open field of the multileaf
collimator' has been given by (among others) Boyer et a1 (1991) and Webb (1991,
1992a,b) (see chapter 2). Boyer et a1 (1991) used an analytic inversion technique
based on deconvolving the point-spread function to obtain photon fluence; Webb
(199 l,1992a,b) optimized a quadratic cost function with prescribed doses to target
volume and organs at risk by the technique of simulated annealing.
7.9. SUMMARY
Historically, some of the earliest approaches to conformal radiotherapy were

based on treatment machines with isotope sources. One such machine, the RFH
tracking cobalt unit, has been in development and use for nearly four decades.
It is still in regular use. Other centres have developed tracking techniques based
around a linear accelerator. Attention now seems to be more focused on shaping
fields with a multileaf collimator. The technology was covered in chapter 5 and
planning techniques were covered in chapter 2. Meanwhile, linear accelerators
have acquired a variety of new features, such as the universal wedge, the
dynamic wedge, independent asymmetric collimators and two-dimensional tissue
compensators. These are sometimes called 'accessories', in that the machine can

References 303
be operated without them, but they are really an integral part of the capability of
the modem machine to perform conformal radiotherapy.
REFERENCES
Bjarngard B, Kijewski P and Pashby C 1977 Description of a computer-controlled

machine Int. J. Rad. Oncol. B i d . Phys. (Supplement 2) 2 142
Blake T, Hickling P A, Lane R A, Nahum A E, Rosen I I and Rosenbloom M E
1987 Assessment of the Philips SL-25 linac for conformation therapy The use
of computers in radiation therapy ed I A D Bruinvis et a1 (Proc.9th ICCRT)
pp 537-540
Boyer A L, Desobry G E and Wells N H 1991 Potential and limitations of invariant
kemel conformal therapy Med. Phys. 18 (4) 703-712
Brace J A 1985 Computer systems for the control of teletherapy units Progress in
medical radiation physics 2 ed C G Orton (New York: Plenum) pp 95-1 11
Brucer M 1956 An automatic controlled pattern cesium 137 teletherapy machine
Am. J . Rontg. 75 49-55
Cheng C W and Chin L M 1987 A computer-aided treatment planning technique
for universal wedges Int. J. Rad. Oncol. Biol. Phys. 13 1927-1935
Chin L M, Kijewski P K, Svensson G K, Chaffey J T, Levene M B and Bjiimgard B
E 1981 A computer-controlled radiation therapy machine for pelvic and para-
aortic nodal areas Int. J. Rad. Oncol. B i d . Phys. 7 61-70
Chui C and Mohan R 1986 Off-center ratios for three-dimensional dose
calculations Med. Phys. 13 409412
Chung-Bin A and Wachtor T 1979 Applications of computers in dynamic control
of linear accelerators Proc. 3rd Int. Computer Software and Applications Conf.
(New York: IEEE) pp 270-275
Davy T J 1985 Physical aspects of conformation therapy using computer-
controlled tracking units Progress in medical radiation physics 2 ed C G Orton
(New York: Plenum) pp 45-94
Davy T J 1987 Conformation therapy methods and systems IAEA-SM-298187
Djordjevich A, Bonham D J, Hussein E M A, Andrew J W and Hale M E 1990
Optimal design of radiation compensators Med. Phys. 17 (3) 397404
ICRU 1976 Determination of absorbed dose in a patient irradiated by beams of
x or gamma rays in radiotherapy procedures ICRU (Washington, DC) report
number 24
Jennings W A 1985 The Tracking Cobalt Project: From moving-beam therapy
to three-dimensional programmed irradiation Progress in medical radiation
physics 2 ed C G Orton (New York: Plenum) pp 1-44
Khan F M, Gerbi B J and Deibel F C 1986 Dosimetry of asymmetric x-ray
collimators Med. Phys. 13 936-941
Leavitt D D, Martin M, Moeller J H and Lee W L 1990 Dynamic wedge field
techniques through computer-controlled collimator motion and dose delivery

Med. Phys. 17 (1) 87-91

Levene M B, Kijewski P K, Chin L M, B j h g a r d B E and Hellman S 1978
Computer controlled radiation therapy Radiology 129 769-775
Loshek D D and Keller K A 1988 Beam profile generator for asymmetric fields
Med. Phys. 15 604-610
Mageras G S , Mohan R, Burman C, Barest G D and Kutcher G J 1991
Compensators for three-dimensional treatment planning Med. Phys. 18 (2)
133-140
Mansfield C M, Suntharalingam N and Chow M 1974 Experimental verification
of a method for varying the effective angle of wedge filters Am. J. Radiol. 120
699-702
McParland B J 1990 The effect of a dynamic wedge in the medial tangential field
upon the contralateral breast dose Znt. J . Rad. Oncol. Biol. Phys. 19 1515-1520
-1991 The effects of a universal wedge and beam obliquity upon the central
axis dose buildup for 6-MV x-rays Med. Phys. 18 (4) 74Cb743
Nakagawa K, Aoki Y, Sakata K, Karasawa K, Muta N, Kojima K, Onogi Y, Hosoi
Y, Akanuma A and Ito M 1987 Dynamic therapy utilizing CT-Linac on line
system The use of computers in radiation therapy ed I A D Bruinvis et a1 (Proc.
9th ICCRT) pp 541-544
Ochran T G , Boyer A L, Nyerick C E and Otte V A 1992 Dosimetric characteristics
of wedges mounted beyond the blocking tray Med. Phys. 19 (1) 187-194
Petti P L and Siddon R L 1985 Effective wedge angles with a universal wedge
Phys. Med. B i d . 30 985-991
Phillips Medical Systems Division 1983 Product data 764 (Eindoven: Philips)
Renner W D, O’Connor T P and Bermudez N M 1989 An algorithm for design of
beam compensators Int. J . Rad. Oncol. Biol. Phys. 17 227-234
Richter J, Klemm P, Neumann M, Nowak G and Sauer 0 1987 Computer control
of a linac for dynamic treatment The use of computers in radiation therapy ed
Saunders W and Chin L M 1986 Innovative techniques: dynamic therapy
and utilisation of non-coplanar beams A categorical course in Radiation
therapy treatment planning ed B R Paliwal and M L Griem (Oak Brook, IL:
Radiological Society of North America) pp 123-128
Schlegel W, Boesecke R, Bauer B, Alandt K and Lorenz W J 1987 Dynamic
therapy planning The use of computers in radiation therapy ed I A D Bruinvis
er a1 (Proc. 9th ICCRT) pp 361-365
Shentall G S and Brace J A 1987 Fully automated planning for conformation
radiotherapy of the spinal axis The use of computers in radiation therapy ed
Tatcher M 1970 A method for varying the effective angle of wedge filters
Radiology 97 132
Thomas S J and Thomas R L 1990 A beam generation algorithm for linear
accelerators with independent collimators Phys. Med. Bid. 35 325-332
Webb S 1991 Optimisation by simulated annealing of three-dimensional

References 305
conformal treatment planning for radiation fields defined by a multileaf

collimator Phys. Med. Biol. 36 1201-1226
-1992a Optimisation of three dimensional treatment planning for volumes
with concave outlines, using a multileaf collimator Proc. ART91 (Munich)
(abstract book p 66) Advanced Radiation Therapy: Tumour Response
-1992b Optimisation by simulated annealing of three-dimensional conformal
treatment planning for radiation fields defined by a multileaf collimator. 2:
Inclusion of two-dimensional modulation of the x-ray intensity Phys. Med.
Bioi. 37 1689-1704
Zwicker R D, Shahabi S , Wu A and Stemick E S 1985 Effective wedge angles for
6 MV wedges Med. Phys. 12 347-349

CHAPTER 8
IMAGING FOR CONFORMAL

RADIOTHERAPY PLANNING
8.1. INTRODUCTION
It is axiomatic that fully three-dimensional images are required for planning

conformal radiotherapy. This restricts the useful imaging modalities to just four:
x-ray CT, MRI, SPECT and PET. Whilst the many other imaging modalities play
major roles in diagnosing disease, they have little if any part in planning conformal
radiotherapy (Webb 1988). Throughout this text we have been referring to the use
of medical images. In order not to detract from the main thrust of presenting
the physics of conformal radiotherapy, the presentation of how these images are
obtained has been left to this point. We now fill in these details, keeping in
mind the application of imaging to therapy. In section 8.2, principles common
to transmission and emission tomography are reviewed. Later sections discuss
some of the important features of these imaging modalities applied to planning
conformal radiotherapy.
Table 8.1 is an attempt to summarize some of the features of the imaging
modalities which play a role in radiotherapy. Most of the questions addressed
cannot be precisely quantified, as values vary from machine to machine. The main
purpose of the table is to show the relative merits of the techniques at a glance and
to provide a feel for how they are used for non-diagnostic purposes.
The simulator plays a major role in ‘conventional’ treatment planning. An
extensive review has been provided by Taylor (1 988). In chapter 1 we have already
discussed how, for conformal radiotherapy, many groups are working towards
using the 3D images in a 3D computer-based planning system to take over many of
the traditional roles of the simulator, including the definition of shaped, possibly
non-coplanar, fields and block design. If this trend continues, the simulator will
become the instrument to check the decisions from such methodology. Simulator
(planar) images will remain important for comparison with megavoltage portal
images (see chapter 6). CT images can also be performed on a simulator and,
whilst this has not yet progressed to generating 3D data-sets, the subject is
reviewed in section 8.8.
306
Principles of imaging by computed tomography 307
Table 8.1. Methods of 3 0 medical imaging4
Method X-ray cr MFU SPECT$ PET
Property being imaged

(A=anatomy; F=function) A A+F F F
Are multisections gathered
simultaneously? (Y=yes; N=no) N Y Y Y
Best spatial resolution (mm) 1 1 9 6
In-plane resolution same ( S ) or
different (D) from slice width? D S D S
Typical data acquisition time
(for multisections) (min) 2 20 15 20
Typical data reconstruction
time (for multisections) (min) 2 2 10 10
Main application site
(H=head; B=body) H+B H+B H+B H+B
Is modality useful for
treatment planning? (C=common;
R=rare; S=sometimes used) C S R R
Is modality useful for
monitoring effects of treatment?
(C=common; R=rare;
S=sometimes used) C S S S
Cost of imaging equipment
(M=moderately expensive;
E=expensive;
V=very expensive) E V M V
How common is equipment in
the UK? (C=common; R=rare) C C C R
Is the physical basis of the
modality still under active
development? (Y=yes; O=already
optimized; C=close to optimized) 0 Y C Y
t This summary does not imply that other methods of medical imaging have
no role in radiotherapy. However, these, even when they are tomographic, are
‘single-section’ imagers and so cannot truly guide 3D conformal radiotherapy.
4 By gamma camera.
8.2. PRINCIPLES OF IMAGING BY COMPUTED TOMOGRAPHY
X-ray CT scanning relies on the principle of reconstruction from projections.

Three-dimensional images may be thought of as a set of stacked two-dimensional
tomograms. The CT scanner records data which can be expressed as line integrals
of the spatially varying x-ray linear attenuation coefficient for a large set of
directions through the slice. The earliest so-called ‘first-generation’ CT scanners

308 Imaging for conformal radiotherapy planning
recorded this data by translating a narrow parallel beam of x-rays impinging on

a detector the other side of the slice (figure 8.1). Each projection comprised
measurements with the line between the source and the detector at a fixed
orientation to the slice. A large set of projections were formed by stepping this
orientation through a range of 180”. ‘Second-generation’ CT scanners were a
variation on this theme, with a narrow fan of radiation falling on a few detectors.
‘Third-generation’ CT scanners had the x-ray beam collimated to a wide fan,
spanning the patient, impinging on a detector array facing the source and co-
rotating with it. Modern ‘fourth-generation’ CT scanners have a ring of detectors
in a complete circle round the slice. The x-ray beam is collimated to a wide
fan, spanning the patient, impinging on the detector. A complete projection
data-set is recorded by rotating the source through 180” plus the fan angle.
The recorded data in all these geometries are proportional to the intensity of
the radiation beam emerging from any particular line through the patient. If
the equivalent measurement is made for a beam passing through air only, the
transmission measurements can be converted to a measure of the line integral of
x-ray attenuation (Webb 1987, 1990a, Swindell and Webb 1988).
Creating the tomographic map of linear attenuation coefficient involves solving
the inverse problem: given the ID line integrals of a 2D spatially varying
property, compute the distribution of that property throughout the section. This
subject is sufficiently important to warrant extensive study and there is an
enormous literature on reconstruction theory (Herman 1980, Natterer 1986).
Arguably the best known method is that of convolution and backprojection (CBP)
(Ramachandran and Laksminarayanan 1971). The ID projections are convolved
with a filter and the convolved data are backprojected along the lines joining
the detection point to the source for each projection element. The task of the
filter is twofold; it compensates for the blurring which would be introduced by
simple backprojection, and it has variable features controlling noise. In the earliest
days of x-ray CT scanning, iterative methods were used (Hounsfield 1968, Gilbert
1972). These started with an initial guess at the reconstructed section and then
refined it until the projections matched the recorded data in some clearly defined
sense. These methods are still used in emission tomography, where an analytic
solution may not be possible. Iterative methods have been substantially developed
beyond the form in which they first appeared in the late 1960’s (Todd Pokropek
1989).
Worldwide there are many well known reconstruction computer packages
which emanate from university departments and are much used by constructors
of special-purpose imaging equipment (Herman et al 1975, 1989, Huesman et
a1 1977). Commercial manufacturers of x-ray CT and SPECT scanners tend to
develop their own in-house reconstruction software, much of it hard-wired and
some of it restricted by commercial secrecies. This is generally considered to
be less of a problem now than it was in the 1970’s, when these disciplines were
relatively young.
To a first approximation, the principles of SPECT are very similar. A gamma

Principles of imaging by computed tomography 309
Figure 8.1. Showing the 'generations' of CT scanner technologies. ( a ) A

1st generation CT scanner. The source S and single detector D face each
other and the line between them translates across the reconstruction circle.
This movement is repeated at each angle in a range of 180". ( b ) A 2nd
generation CT scanner. The source S emits a narrow fan of x-rays in the
plane of the section which impinges on a small detector bank DB. The source
and the detector bank translate together across the field for a 180" range
of angles. The angular sampling can now be coarser than in arrangement
(a), reducing scanning time. ( c ) A 3rd generation CT scanner. The source
emits a wide fan of x-rays in the plane of the slice which impinges on a large
bank of detectors DB opposite. The source and coupled detector bank rotate
together around the reconstruction circle. The angular coverage need only be
+
180" the fan angle, although there is no reason why more datafiom a full
circle could not be used to improve image quality. The major disadvantage of
this geometry is the generation of circular artefacts. ( d )A 4th generation CT
scanner. A s in the 3rd generation CT scanner, the source S emits a widefan of
x-rays and rotates through at least 180" +the fan angle. This time the detector
ring DR is static and, as the source rotates, direrent detecting elements are
called into play. This eliminates circular artefacts.
camera rotates around the patient through 360", recording at fixed orientations a
planar image into a square digital matrix (figure 8.2). Each line of data, in the plane
of rotation of the camera, for each planar image is a 1D projection of the activity in
the corresponding slice. The situation is actually rather more complicated than this
because the line integral really records attenuatedactivity and special steps have to
be taken during reconstruction to cope with this. Once again, a range of geometries
is possible. The most common method has a parallel-hole collimator attached to
the camera and data are recorded analogous to 'first-generation' CT (without the
need for translation, of course). Fan-beam collimators mimic 'third- or fourth-
generation' CT. There is no emission equivalent of 'second-generation' CT. A
single rotation of the gamma camera records all the data needed for a complete

3 10 Imaging f o r conformal radiotherapy planning
Figure 8.2. Showing the principle of SPECT. The patient lies horizontally. If
the images are to be correlated with x-ray CT and MRI images, the couch must
be $at. A gamma camera rotates around the patient through 360°,usually
in a series of discrete angular increments. At each orientation the data are
recorded into a square matrix as shown. (The camera is usually circular and is
not shown for clarity.) One particular line of data on the camera in the plane
of rotation (shown as a solid line) contributes to the reconstruction of one
particular slice in the patient (as shown). Unlike x-ray CT, all the projection
data for a series of slices are recorded in one single rotation of the camera.
set of tomographic slices. Data are generally reconstructed into transaxial cross
sections and re-ordered into sagittal and coronal slices. Emission SPECT is much
slower than transmission CT. Typically 15 m may be needed for a complete study,
whereas each x-ray CT set of data are taken in a few seconds (Webb et a1 1985a).
Positron emission tomography can operate on similar principles of reconstruc-
tion. When a positron emitter decays, two 5 11 keV gammas are emitted more or
less back-to-back. They are detected in coincidence, thus defining the line along
which the decay must have occurred (figure 8.3). Some PET systems have a ring
of scintillation detectors surrounding the patient and the recorded data can be ar-
ranged into projections at a large set of orientations, from which the emission to-
mogram may be recorded in the same way as a SPECT image. Other PET systems
make use of rotating area-detectors. For these it is more appropriate to backproject
each recorded coincidence event into the reconstruction space to create a blurred
3D image. The point-spread function of the imaging system is then deconvolved
to produce a sharp image. In this deconvolution stage, appropriate windowing
can control the noise. The usual trade-off arises between controlling the effects of
noise and the spatial resolution obtainable (Phelps 1986).
Although magnetic resonance imaging, following exactly the principles of x-
ray CT, can be performed, until recently this method has been thought too slow
to be practical. Hence MRI tomograms are generally recorded differently. NMR
signals are radiofrequency emissions resulting from the relaxation of spins which
have been excited by radiofrequency pulses. The exact location from which each
signal arises is determined by the frequency of emission, which in turn is made
to vary spatially over the region to be imaged by applying magnetic gradients in
several orthogonal directions superposed to the main, large magnetic field (of the

X-ray computed tomography 3 11
Figure 8.3. Schematic showing the principle Of PET scanning. The patient is
surrounded by a ring of detectors (only a few are shown) collimated to a thin
plane parallel to the plane of the paper. Coincidence circuits (not shown) link
the detectors. A region in the brain of the patient (shown dark) is emitting
positrons and three coincidence paths are shown. (In practice coincidence
paths emanate in all directions in the plane.) A single tomographic plane is
reconstructed from this data. To obtain more planes, more coincidence rings
are required. Alternatively, ‘area detectors’ may be used.
order 2 T). MRI tomograms can be made to reflect different physical properties by
varying the ‘pulse sequences’ which capture the data. This is a complex subject
and the above is a big simplification (Chen and Hoult 1989).
There are many physics and engineering design considerations in optimizing
these medical-imaging modalities and whole volumes have been devoted to each.
It would not be appropriate here to oversimplify these important considerations
and readers wishing to know more might consult one of the following: Moores
et al (1981), Hamilton (1982), Wells (1982), Gifford (1984), Guzzardi (1987),
Aird (1988), Webb (1988), Chen and Hoult (1989). The next sections focus on
some of the aspects of these imaging technologies which are particularly relevant
to conformal radiotherapy.
8.3. X-RAY COMPUTED TOMOGRAPHY
X-ray computed tomography has a long and complex history (Webb 1990a).
However, the important event for clinical practice occurred in April 1972 when
Sir Godfrey Hounsfield unveiled the world’s first commercial CT machine (from
the EM1 Company) at the 30th British Institute of Radiology Annual Congress
(Hounsfield 1973). It was not long before CT images were routinely forming the
basis of radiotherapy treatment planning. The present state of the art was reached
with amazing rapidity by the mid to late 1970’s (Susskind 1981, Blume 1992),
and whilst small improvements continue to be made to CT scanners, no further
quantum leaps are expected.
The main requirements for therapy CT scanning include the need for the patient
to lie on a flat couch in the treatment position so that planning is based on the

3 12 Imaging for conformal radiotherapy planning
Figure 8.4. Schematic showing how a topogram isformed. The patient lies
on afiat couch and is translated through the beam, which is collimated to a
narrowfan. In thepicture the beam is in either of twopositions, labelledas 0‘
and 90”. A digital radiograph is formed which shows dotted lines indicating
the location of the CT slices. (From Krestel(l990).)
geometry as it will be at treatment time. The patient quietly respires, and time
taken to assure patient comfort is time not wasted in this respect. Radio-opaque
markers are placed on selected skin tatoos to act as landmarks. A ‘scout view’ (or
‘topogram’) is recorded to determine the necessary range of travel of the couch.
This is formed by taking the patient through the CT scanner with the fan beam
and detectors stationary, thus generating a kind of planar radiograph (figure 8.4).
This has the characteristic feature that the beam diverges normal to the direction of
travel and is parallel in the orthogonal direction. In this sense, such a radiograph
could not be recorded by a 2D detector and an unmoving x-ray source (figure 8.5).
Sections are taken from a few cm below the target volume to a few cm above and
registered on to the ‘scout view’. These sections are generally spaced 5 mm apart
in the body, but may be as close as 2 mm in the head or in the periphery of fields in
the body where beam penumbra will occur. Contrast media are used for selected
examinations. Reconstructed CT numbers convert to tissue electron densities.
Image data are generally transferred to the planning computer by magnetic tape.
There have been countless reviews of the impact of x-ray CT on therapy
planning. The issues are discussed well by Goitein (1983) and Hogstrom (1983).
Reviews by Webb (1990b) and Dobbs and Webb (I 988) included a discussion of
special-purpose CT scanners (see also section 8.8).

Magnetic resonance imaging 3 13
Figure 8.5. Topogram to facilitate the localization of CT slices within the

torso. The dotted lines give the slice numbers. (From Krestel(l99O)J
8.4. MAGNETIC RESONANCE IMAGING
At first encounter, magnetic resonance tomography and x-ray computed

tomography appear to have similar roles. They both produce high-resolution fully
three-dimensional tomographic data-sets. Both have transformed the practice of
diagnostic radiology. However, important differences between them ensure that
they play complementary rather than competing roles in diagnostic radiology. CT
had not been in use for diagnosis for long before the images were being exploited
to assist radiotherapy planning. Hence, when diagnostic MRI became widely
available, the same questions were asked concerning its suitability for assisting
radiation therapy planning. Early experience suggested the disadvantages in MRI
were insurmountable (Henkelman et a1 1984, Sontag et a1 1984). This view has
been modified with the passage of time. Experience has shown that, for this
application also, the role of MRI is complementary to that of CT (Coffey et a1
1984, Ten Haken et a1 1991).
There are important differences which ensure that this complementarity is
likely to remain. Whilst both modalities have good spatial resolution, MRI scan
times are much longer than for CT. Hence there is a danger of artefacts from
patient movement and also reduced patient throughput. MRI cannot give electron-
density images needed for x-ray therapy planning, whereas the electron density of
tissues directly follows from the image of CT numbers representing x-ray linear
attenuation coefficient. In its favour, MRI directly generates non-transaxial data-
sets, including sagittal, coronal and oblique. X-ray CT is geometrically very
accurate, whereas image distortion can be a problem for MRI. High-sensitivity
MRI sometimes makes use of head and body coils which restrict access to and
positioning of the patient. In its early days, x-ray CT had a limited patient aperture,
but this is less of a problem with the latest CT machines. For both modalities the

Figure 8.6. Brain metasrasesfram a primary breast tumour which are not
visible on the C T s k e (left). URI show the metastases with high tumour-specifrc
contrast on both a T2-weighted sequence (centre), and to a lesser extent on a
TI-weighted sequence (right). In the central image the lesions show as bright
spots; in the right-hand image they show as dark spots. (From Henkelman
(1992).) (Reprinted with permission fiom Pergamon Press Ltd, Oxford, UK.)
patient should be imaged lying on a flat couch with quiet respiration. Finally,
one of the strengths in diagnostic work, the ability of MRI to image different
physical properties within any slice by varying the pulse sequences, might be
thought a limitation for its use in therapy planning. The MRI images may become
too dependent on the choice of acquisition parameters, whereas there is a certain
inherent consistency in how CT images are formed (Fraass et a1 1987, McShan
and Fraass 1987).
The ability of MRI to display functional differences between benign and
malignant tissue, which often extend beyond the anatomical differences imaged
by x-ray CT, is the driving force behind efforts to harness it into the planning
process. For example, figure 8.6 shows brain metastases from a primary breast
tumour which are invisible on a CT scan but show up on MRI. Tumours can have
characteristically longer relaxation times than the corresponding normal anatomy,
providing the physical basis for tumour contrast. Sometimes the MR sensitivity
to flow can be used to generate image contrast (Henkelman 1992). Shuman er a1
(1985) reported a study of 30 patients who were imaged with both CT and MRI.
Radiation therapy planning was initially based on the CT data alone and the effect
of having the additional MRI information was systematically assessed. In over
50% of the cases the radiation therapy plan was changed as a result of having
the MRI data. In many other cases, whilst no changes were made, confidence was
increased in the CT-based plan. Whilst the authors of this study admitted that there
might be some element of bias in the study, in that patients were selected where
MRI data might be expected to provide additional information, an alternative way
of looking at this situation is that the new (MRI) imaging modality was being
used appropriately. MRI can also be a useful tool for monitoring the efficacy of
therapy by imaging changing tissue function. Indeed Henkelman (1992) argues
that we should not expect further great improvements in geographical definition
of structures, but look to improvements in monitoring physiological function.

Magnetic resonance imaging 3 15
The literature on the use of MRI in radiotherapy planning largely concentrates

on how to overcome the difficulties (Henkelman et a1 1984). One of the main
criticisms of the use of MRI in radiotherapy planning is that images may be
distorted. Since conformal therapy requires exquisite precision, this feature
appears contrary to aims. The distortion may arise in one or both of two
ways. Geometric distortion may arise from inhomogeneities in the magnetic field,
particularly at the edge of body images. This is a smaller, if non-existent, problem
in head scanning. Further distortions may arise due to the patient perturbing the
field. The question arises whether these distortions are constant with time and
from patient to patient. If not, individual corrections are required. Phantoms of
known geometry, e.g. a matrix of fluid-filled tubes, can be imaged to characterize
the distortion from the former cause. Unwarping algorithms may then restore the
geometric consistency of the data (Fraass et a1 1987). With long scan times, patient
movement may lead to artefacts and/or distortion.
MRI and CT images require to be registered so that the electron density data from
CT can be mapped on to the corresponding MRI images where the target volume
may be more easily determined. Henkelman et af (1984) showed a complete lack
of correlation between electron density and MRI image data; registering MRI and
CT is the way to overcome the problem. It may not be possible to immobilize
patients identically for both imaging sessions (e.g. because of the coils mentioned
earlier). CT immobilization methods involving ferromagnetic materials are also
excluded for MRI. Hence the many registration techniques discussed in chapter 1
are called into play. In particular, the use of extemal markers can form the basis
of correlation. Fraass et af (1987) comment that a source of inaccuracy arises
because of the slippery nature of the oil-filled markers used in MRI. Also, these
need to be arranged so that they show as points in all sections, including those
which are non-transaxial. Figure 8.7 illustrates this.
After registration, images are presented to the 3D planning system. A difficulty
which may arise here is that the pulse sequences for orthogonal slices may have
been different, leading to structures appearing with different grey levels in each
slice. When these are merged into an axonometric display, the effect may be
confusing. To overcome this, some workers prefer to show MRI and CT data
separately, but alongside each other, and transfer structures from one data-set to
the other via linked cursors and similar tools.
Three-dimensional planning systems, traditionally CT-based, may not be able
to accept 3D MRI data, especially from non-transaxial planes. Some systems, such
as UMPLAN, are specifically designed to use MRI and CT data interchangeably.
The beam algorithms are written to yield the dose distributions in such planes and
to display them accordingly.
Bearing in mind these difficulties, MRI is least problematic in the head, where
distortions may be minimal, and MRI is indicated because of its ability to show
the tumour and oedema better than x-ray CT. MRI presents a particularly strong
case for routine quality assurance.
In summary, MRI is not yet routinely used for planning radiotherapy, but this is

A B
I \ morhers
\ I
1
C T MARKERS MR MARKERS
' 8
/i ' Transverse
Transverse marker Special
markers J Sogi t tal
marker
1 U
I A X I A L MR
S A G I T T A L MR
Figure 8.7. ( A ) shows three line markers used in a multisection CT study.

These will show as dots in transaxial tomogram. ( B ) shows how orthogonal
line markers may be used in MRI, showing as points in sagittal dices (such
as (C)) and either points or a line in transaxial slices (such as (D)). The
intersections between the lines can define points in the two data-sets, which
may be used for image registration. (From Fraass et al(1987).) (Reprinted
with permission @om Pergamon Press Ltd, Oxford, UK.)
expected to change with the passage of time.
8.5. ULTRASOUND IMAGING
Despite its enormous importance for diagnosis, ultrasound imaging plays virtually
no part in planning conformal radiotherapy. Ultrasound is reflected at barriers
of high acoustic-impedance mismatch. It is difficult for ultrasound to penetrate
bone or gas, so ultrasonic imaging is unable to generate 3D tomographic data-
sets for treatment planning. Ultrasound has been used to determine chest wall
thickness for breast treatment planning (Lock and Coules 1986, Taylor 1988).
Comparison of the extemal contour and lung position located by ultrasound with
CT and a life-size pantograph showed that the contour points from the ultrasound
scan were never more than 3 mm from the corresponding points determined by
the comparative method. The ability to obtain a contour in a non-transaxial plane
was an advantage for ultrasound.

Positron emission tomographqLPET 3 17
With the passing of the old compound ultrasound imaging systems and the
emergence to total predominence of the hand-held real-time scanners, any last
semblance of a coordinate registration system has been lost. Unless these features
were added to modem ultrasound scanners it would not be possible to use them
for planning in the same way as CT or MRI. This is not impossible, but it has not
been done.
This situation may change soon, since there is currently considerable activity in
3D ultrasound data collection. Scanners which automatically generate a sequence
of parallel-slice images over a pre-defined volume in about 5 s have been designed.
Early commercial versions already exist. Research programmes are studying
options for real-time 3D scanning using parallel processing systems. These will
still be restricted to applications where the presence of air and bone are not
limitations (Bamber 1992).
8.6. SINGLE-PHOTONEMISSION COMPUTED TOMOGRAPHY4PECT
Both SPECT and PET provide three-dimensional functional images (Ahluwalia et

a1 1989). Since functional defects may not necessarily correlate with anatomical
defects, images can be useful for planning therapy. In chapter 1 we looked
at methods by which SPECT and PET could be registered and entered into the
planning process. The use of emission tomography to assess the effectiveness
of radiotherapy is equally important and calls for the same precision in image
registration. It is required to superimpose three-dimensionaldose distributions on
to three-dimensional functional images.
SPECT is characterized by a fairly poor spatial resolution if a parallel-hole
collimator is used (Webb et a1 1985a). This depends on the reconstruction filter,
but might be typically as poor as 1.5 cm. For correlating body images, a flat
couch should be used during scanning. Since scan times are typically 15 m,
the patient is not suspending respiration, so there is the possibility of generating
artefacts in thoracic images. SPECT images are generally reconstructed into cubic
arrays with cubic voxels, since the planar images, taken as the camera rotates,
are square with square pixels. The main advantage of SPECT is that a wide
variety of radiopharmaceuticals may be imaged (Ott et a1 1988). Some ‘single-
section’ emission tomographic devices can provide images with improved spatial
resolution, but, unless the patient is translated through the machine to obtain
multiple slices, these images would be less useful for assisting conformal therapy.
8.7. POSITRON EMISSION TOMOGRAPHY-PET
Positron emission tomography opens up the possibility of imaging radiolabelled

elements naturally ocuning in the body, such as hydrogen, oxygen and carbon.
For those radionuclides with a short half-life, an on-site cyclotron is essential and

many ‘PET centres’ have the cyclotron and imaging technology close to each other.
However, PET without a cyclotron is possible using, for example, “Ga-labelled
radiopharmaceuticals.
The spatial resolution of PET images is generally rather better than SPECT
images. Precise values depend on the radionuclide and on the method of detecting
the photons, but spatial resolution of the order 6 mm is possible. Machines
fall into two categories: ‘ring systems’ in which there are a number of rings of
scintillation detectors and ‘area-detector’ systems. The former produce exquisite
functional images, but with a small number of slices covering a limited axial
extent. The latter generate as many slices as a SPECT system, covering a larger
axial extent; in principle a bonus for assisting conformal radiotherapy (Ott et a1
1985). Research in imaging technology is still very active, with the supporters
of area-detector systems working hard to emulate the good sensitivity achievable
with ring systems.
8.8. CT IMAGING ON A RADIOTHERAPY SIMULATOR
A number of groups worldwide have investigated the extent to which the imaging
requirements for radiotherapy planning may be met by the development of non-
standard CT scanners. Some of these groups have constructed apparatus based on a
radiotherapy simulator gantry. Others have built special-purpose CT devices with
radioactive sources, but these will not be reviewed as in general they were early
developments not intended for clinical use. The major use of simulator-based
machines has been in planning radiotherapy in the upper thorax, where tissue-
inhomogeneity corrections are important.
After the first commercial CT scanner was announced in 1972 (New Scientist
1972, Ambrose and Hounsfield 1972), it was not long before body scanners were
developed (Ledley et a1 1974, Ledley 1976). Both head and body scanners were
used to generate ‘therapy CT scans’, images of internal tumours and anatomy,
useful for radiotherapy treatment planning (Parker and Hobday 1981, Dobbs and
Webb 1988).
Full 3D diagnostic-quality CT data-sets are essential for planning most
conformal radiotherapy. However, a small body of people developed simpler CT
machines to assist with specific treatment planning problems and these are briefly
reviewed here. Planning to conform the high-dose volume to the breast, avoiding
irradiating too much lung, is one such application. A full review has been written
by Webb (1987, 1990b). If multi-section CT could be achieved with a simulator-
based scanner, this would be very helpful to 3D planning conformal therapy and
greatly speed up patient throughput.
8.8.1. Systems for non-diagnostic computed tomography (NDCT)

The term NDCT is here used to mean all those singie-slice computed tomography

CT imaging on a radiotherapy simulator 3 19
machines whose aim is to produce CT cross sections of lower quality-

i.e. poorer spatial and electron density resolution-than obtained by state-
of-the-art commercial CT scanners. The latter were of course developed for
diagnostic purposes, their use in treatment planning being inspired lateral thinking
subsequently. An abundance of terms appears in the literature for NDCT machines,
including ‘simple CT scanners’, ‘altemative CT scanners’, ‘non-commercial CT
scanners’ etc. The term NDCT is prefered because it immediately differentiates
the purpose of such machines and is a valid generic term. It removes the sense
that NDCT scanners are in some way inferior. They can and do provide enough
data for the purpose for which they were designed and engineered.
Most NDCT machines were designed and built in university and hospital
departments, rather than by commercial companies. Hence, in reviewing their
construction and capabilities we are faced with a miscellany of technology. No
two machines are alike, although some use common detector technology.
8.8.2. NDCT on a simulator with an image intensifier as detector

A number of workers built NDCT systems based around the use of an image
intensifier as x-ray detector. The idea appears to have been first suggested in a
German patent by Dummling (1970). Baily (1977,1979), Baily et a1 (1976a,b)
and Kak et a1 (1977) were among the first to investigate this possibility, coupling
an image intensifier to a vidicon TV tube and recording real-time fan-beam x-
ray projections. The method took account of the non-linear logarithmic response
of the image-intensifier detector. The data were obtained in fan-beam geometry
by collimating the x-ray source to a narrow fan spanning just eight TV lines. At
the time of this work (circa 1975) reconstruction from fan-beam projections had
received little attention and Kak et a1 (1977) decided to treat the beams as if they
were parallel.
Harrison and Farmer (1976, 1978) in Newcastle used a similar technique to
image body cross sections. They took the work into the clinic, based around
a radiotherapy simulator and were well aware of the potential for improving
radiotherapy planning (Harrison 1981). A 48S2 cm2 fluorescent screen was
viewed by a low light-level Intensifier Isocon television camera. This detector
was mounted on a TEM Ximatron Mark 1 gantry. Image data were reconstructed
off-line by an IBM-370 computer using a hybrid convolution and backprojection
(CBP) algorithm. This treated the projection data as if collected in parallel
geometry for the convolution part but then backprojected along the correct fan-
beam lines. The early papers from this group noted that the achievable spatial
accuracy of some f 5 mm was consistent with getting dosimetry right to within
2-5% for 6oCo,4 MV and 8 MV x-radiotherapy.
The NDCT system at Newcastle is one of the few examples of a system
which has been heavily used clinically (Kotre el a1 1984, 1986, Kotre and
Harrison 1988) for both planning external radiotherapy of thoracic regions
and assisting localizing the sources in brachytherapy. It does suffer from the

<KAY SOuRCE
TABLE
TOP
Figure 8.8. The 'half-fan' method of projection data capture with an offset
image intensifier; as used by several groups. The source and detector must
make a complete 360" rotation in order to capture the data. The intensifier
is offset to increase the field of view, The wedge close to the x-ray source is
to compensate (approximately)for the changing shape of the patient so that
the data do not have too large dynamic range. (From Redpath and Wright
(198S).)
major disadvantage of poor field of view (30 cm diameter reconstruction circle)

necessitating combining NDCT data with conventional lead-wire outlining for
large patients. This is because the image intensifier is not offset to one side as
in some other developments.
A team from Edinburgh overcamethe field-of-view limitation of body-section
imaging using a radiotherapy simulator by introducing a lateral offset for the
position of the image intensifier of some 15 cm at a distance of 35 cm from the
isocentre (figure 8.8). This 'half-fan mode' of data capture necessitates a full
rotation through 360" followed by combining half projections into half the number
of full projections (Redpath and Wright 1985, Redpath 1988a,b,c, Wright et a1
1984). It was found most convenient to re-bin the fan-beam data into equivalent
parallel projections prior to reconstruction.
The use of a television system coupled to a fluorescent screen and image
intensifier on a rotating gantry proved a popular way of creating a NDCT
scanner.As early as 1978, one company, Oldelft (N V Optische Industrie 'de Oude
Delft', Delft Holland) produced a commercial machine (Duinker et a1 1978). This
was subjected to a rigorous experimental investigation by Kijewski et a1 (1984)
in Boston. The scanner was utilized in two quite separate modes. Firstly, it was
used, as the manufacturer intended, to generate images by analogue reconstruction
means. Secondly, the projection data were digitized and reconstructed on a digital
computer. The images obtained in the two ways were compared quantitatively.
The data capture system comprised a flat fluorescent screen of gadolinium
oxysulphide coupled by mirrors and lenses to an image intensifier and an Isocon

CT imaging on a radiotherapy simulator 321
television. For analogue mode, the detector was offset to collect 1500 half-
projections in 360" (reconstruction circle 40 cm diameter). For digital mode the
offset was eliminated and the reconstruction circle shrank to 20 cm diameter. The
slice width for the analogue mode was 7 mm (16 TV lines) and 3.5 mm (8 TV
lines) for the digital mode, Digital projection data were obtained (143 frames in
360") using a Colorado Video Corporation model 270A video digitizer interfaced
to a VAX 11/780 computer, the digitization taking 2.5 h. Digital reconstruction
used a fan-beam CBP algorithm with a ramp-Hanning windowing filter.
The results of experiments carried out both ways showed that the spatial and
density resolutions of the analogue images were 3.1 mm and 3.5% respectively.
The corresponding figures for the digital reconstruction were 1.1 mm and
15.4%. Naturally these latter figures could be changed by implementing a
different reconstruction filter. Kijewski et a1 (1984) found that when a blurring
filter was chosen which yielded the same (3.1 mm) spatial resolution as for the
analogue method, the noise reduced to 3.2 %. Further, they noted that analogue
reconstruction made use of some 10 times the number of projections and that had
this number been used for digital reconstruction the noise value would have fallen
even further to 1.8% at this spatial resolution.
At London's Hammersmith Hospital a system was constructed using half-
fan-beam projections recorded by a Siemens Siretom RBV 30H triplex image
intensifier with a caesium iodide screen of curved cross section coupled to a
Siemens VideomedN vidicon TV camera (Amot et ai 1984). These were mounted
on a TEM Mark 5 Ximatron gantry. A distinguishing feature of this development
was the attention paid to designing a hardware electronic filtering circuit which
approximated to the SheppLogan filter. Spatially filtered data were backprojected
in a fast-arithmetic array processor (Cynosure Imaging Systems Inc). This filter
used only nine data points (four each side of the element being filtered) and was
implemented in a binary-word shift register with bit shifting and addhubtract
circuitry. The filtering took < 40 ms per projection and so the reconstruction
became essentially real time.
This group corrected for the problems of image intensifier non-uniformity and
the effect of the earth's magnetic field. They concluded that no figure for spatial
resolution could be given because the image of a fine wire vaned with position in
the field of view. Density resolution was of the order 3-5 %. The performance of
the system was satisfactory for delineating skin contours, bony tissue, abnormal
and normal lung fields and was thus suitable for localization and inhomogeneity
correction for radiotherapy. It would appear however that this group had loftier
hopes because they found that soft-tissue discrimination was not acceptable and it
was, for example, not possible to differentiate ventricles nor tumour in the brain.
Figure 8.9 shows a head and chest transverse section to illustrate the quality of
data from this machine.
Cynosure Imaging Systems Ltd went on to develop their array processor so that
analogue television signals from a rotating image intensifier could be digitized
in the AP and then reconstruction took place by digital convolution and digital

Figure 8.9. Images of transverse sections of ( a ) head and ( b ) chest of

normal subjects taken at nominal 125 kVp and 4.2 mA on the Hammersmith
Hospital system, which utilized an image intensifier detector. (From Arnot et
a1 (1984).)
fan-beam backprojection. The whole process took only 2.5 s (Cynosure 1985).
One problem with all such systems based on an image intensifier is the scatter
contribution to the image. The systems did not employ collimation at the detector
(unlike commercial DCT systems) and this was a limitation.
Very recently a commercial system (the Ximatron CT option) has been marketed
by Varian Associates (Varian 1992) (figure 8.10). The detector is an offset
image intensifier with photodiode outrigger. A collimator at the detector assists
scatter rejection. 500 projections are recorded with 512 elements each. Data are
reconstructed by a 2D fast Fourier transform method into an image matrix of 512*
< <
pixels. A spatial resolution of 2 mm and density resolution of 1% is achieved.
This is very close to the performance of a diagnostic CT scanner. It takes 1 minute
to take the data and a further 1.5 minutes to reconstruct the image (figure 8.11).
The reconstruction circle is 50 cm in diameter and the aperture is 80 cm.
8.8.3. NDCT with multiwire or scintillation detector

A group from the University of Califomia have reported two machines. The first
(Smith et a1 1976a,b, 1977, 1978, 1980)employed a multi-elementxenon detector
replacing the image intensifier on a radiotherapy simulator. This detector had a
collection efficiency of 99% at the energy in question, comprised 121 radially
oriented wires subtending a 40" arc at the x-ray source. Reference detectors were
used to compensate for the fluctuating x-ray intensity and software normalization
between channels ensured that the output from each channel was stable. The
projection data (typically 180 views) were re-ordered into parallel geometry and
reconstructed by CBP algorithms employing either the Ram and Lak filter or
the Shepp and Logan filter. Spatial resolution of some 7.2 mm and density
resolution of 4.1% were achieved. Live subjects were scanned, but the small
patient aperture of only 30 cm limited the device principally to head scanning,

Figure 8.10. The geometry of the Varian CT simulator. The image intensifrer
is offset and there is an additional post-collimation extension detector
(outrigger). At the x-ray head is a precollimator, with a reference detector
and a 'bowtie' filter to reduce the dynamic range of the output values. The
image intensifrer is coupled to a camera which is differentfrom the one used
for routine screening. The source and detector must make a complete rotation
about the patient to capture the data. (From Varianproduct literature.)
Figure 8.11. CT scan of a head taken with the Varian CT simulator with
a magnifying glass, showing detail of the sinus. (From Varian product
literature.)

although the possibility of improving breast radiotherapy by imaging in the ‘arms

up’ position was considered. A second limitation was the rather large dead space
(some 7 mm) between detector elements.
Smith et a1 (1982) built on their earlier experience to construct a simulator-
based CT scanner for radiotherapy planning. They used a bank of 256 solid-state
cadmium tungstate crystals coupled to silicon photodiodes and mounted these on
a separate C-arm gantry which sat at the opposite end of the simulator couch from
the simulator gantry. Using data capture electronics with 17-bit dynamic range,
their system performance (4% density and 2.3 mm spatial resolution) was closer
to that of a diagnostic CT scanner than to most other NDCT scanners reviewed
here.
8.8.4. The Royal Marsden Hospital CT Simulator

The author and colleagues worked on the development of NDCT scanners between
1976 and 1987. The work progressed in three main stages, the first two being test-
bed arrangements leading to the development of the RMH ‘CT Simulator’.
Work began in 1976 on a system for recording transmission projection data on
film in a special cassette holder which locked on to (and thus replaced as detector)
the image intensifier of a radiotherapy simulator (Webb et a1 1977a,b, 1978, 1981,
Webb and Leach 1981).
This group also constructed a NDCT scanner based on a modified J and P
Engineering gantry and having a I3’Cs source (Webb et a1 1981, Webb and Leach
1981, Leach and Webb 1979, 1980, Leach et a1 1982, 1984a, and Flatman et a1
1982).
The instrument designed for patient scanning (the ‘CT Simulator’) arose from
these studies. Construction began in 1980 and was completed in 1983. (Bentley
et a1 1987, Dobbs and Webb 1988, Leach and Webb 1982, Leach et a1 1984b,
1985a,b, Maureemootoo et a1 1988, Toms et a1 1986a,b, Webb 1982, Webb et a1
1984a,b,c, 1985b, 1986, 1987a,b, 1988, and Webb and Long 1985).
The machine comprises a detector of novel design, a bar of plastic scintillator
viewed at each end by photomultiplier tubes. The detector is collimated first by a
narrow slit (3 mm wide), to define the section to be reconstructed from the x-ray
fan-beam projections. Additionally, the plastic scintillator of circular section is
collimated by a pair of rotating lead cylinders, one inside the other and rotating
in the opposite direction to each other. Cut into these cylinders are helical slots,
also 3 mm wide and spiralling in opposite ways. The effect of rotating these two
helical slots is that a diamond-shaped aperture to the scintillator ‘scans’ from one
end to the other, once per rotation of the cylinders. As soon as the aperture reaches
one end, it ‘reappears’ at the other end. The detector collimators rotate once
every 3.31 s continuously whilst the gantry supporting the detector and the source
rotate continuously through 360” in about 5 minutes. By sampling the summed
output from the photomultiplier tubes every 100 ms, ‘projections’ (of length 331
elements-but see below) are continuously recorded. The novel collimation thus

Figure 8.12. View of the Royal Marsden Hospital ‘CT Simulator’. This was
builtfrom a Fairey Engineering simulator with excellent geometric bearings.
The source collimation was re-engineered and the special one-dimensional
detector was added opposite the source. A completely new isocentric couch
facility was engineered.
gives the detector its positional sensitivity. Additionally, the detector is a photon
counter rather than a recorder of electrical current.
This is not an efficient method of collecting data because the x-ray beam is
not steered with the moving aperture. The method was adopted because the
technology of photon counting was well known and reliable and the detector
was inexpensive to construct. The delivered dose, despite being largely wasted
dose, is not prohibitive for patients receiving radiotherapy. The projections
recorded in this way are in a unique geometry. Instead of being able to think
of a small finite number of projections at some angular orientation of the gantry
with many projection elements, one needs to think of projections as being only
single elements but at an enormous number of orientations. The problem of
reconstructing this data was solved by Herman (1982) who showed that the
data could be reorganized into an equivalent small number (of the order 100) of
‘conventional’ fan-beam projections.
The scanner is controlled by a PDPllO5 computer through a NASCOM Z80
micro, data being captured direct to disc. The data are then transferred via floppy
disc to a VAX 750 for image reconstruction. Typically 4.1 mm spatial and 13%
density resolution is achieved with a dose to the sternum of 55 mGy. The slice
width is 4.6 mm. The scanner is shown in figure 8.12 and the detector in close up
in figure 8.13. The patient aperture was some 95 cm and the reconstruction circle
46 cm, entirely adequate for even the largest patient.
The RMH CT Simulator was used to study the importance of including cross-
sectional imaging data in the treatment planning process for post-operative breast
radiotherapy. It was soon recognized that images taken at the mid-line, upper

326 Imaging f o r conformal radiotherapy planning
Figure 8.13. View of the detectorfor the RMH a-Simulatorfrom the source
side. The rotating lead cylinders visible infigure 8.12 are below the extra slit
collimator. The two mu-metal-shieldedphotomultiplier tubes are in the brass
tubes at each end of the cylindrical plastic scintillator.
and lower borders of the treatment field were very different with respect both
to extemal contour and internal lung spaces (see figure 8.14). Conventional
treatment planning based on simple extemal contour localization by lead-wire ran
the risk of compressing the breast, as well as not providing for accurately tailoring
the treatment to the individual chest-wall thickness and lung size. The study
showed that significant under and overdosage to breast tissue could occur in these
circumstances, that lung tissue was being unnecessarily irradiated by conventional
non-conformation techniques and that both could have serious effects on local
control in the breast and pulmonary complications.
8.8.5. Cone-beam x-ray CT on a simulator

If CT performed on a simulator is ever to become a true contender for assisting
planning conformal radiotherapyfor sites other than the breast, multiple-slice data
must be obtained. Presently it takes of the order of a minute to image one slice,
limited by the speed of rotation of the gantry. One might envisage an area detector
which would allow more slices to be simultaneously imaged, but as yet this has
not been engineered, although reconstruction techniques are well understood for
this geometry (Feldkamp et a1 1984, Webb et a1 1987~).

Summary 327
Figure 8.14. CT images of the breast (with the patient in the treatment
position) taken with the RMH CT Simulator (a) mid-field, (b) upper border,
( c ) lower border. This illustrates the large changes in both internal and
external contours across the radiation field and indicates the need for fully
three-dimensional imaging to be engineered on a simulator. As yet no one
has achieved this.
8.9. SUMMARY
Medical imaging ‘drives’ conformal radiotherapy. For true 3D conformation, only

those imaging modalities which are truly three-dimensional can play a major role.
These are x-ray CT, MRI,SPECT and PET. There are special imaging requirements

when these are used to assist radiotherapy, rather than for diagnosis. Imaging
serves a dual role: planning and monitoring treatment. For both it is necessary
for 3D image data-sets to be registered so regions of interest in one modality
(for example x-ray CT with a superposed dose distribution) can be transferred
to another modality (for example PET monitoring changed function). Because
of the importance of the simulator for planning, the development of x-ray CT on
a simulator may play an increasing role, but single images are not enough and
a digital area detector is needed to generate 3D images with the patient in the
treatment position.
REFERENCES
Ahluwalia B D (ed) 1989 Tomographic methods in nuclear medicine: physical

principles, instruments, and clinical applications (Boca Raton: Chemical
Rubber Company)
Aird E G A 1988 Basic physics for medical imaging (Oxford: Heinemann)
Ambrose J and Hounsfield G N 1972 Computerized transverse axial tomography
Brit. J . Radiol. 46 148-9
Amot R N, Willetts R J, Batten J R and Orr J S 1984 Investigations using an x-ray
image intensifier and a TV camera for imaging transverse sections in humans
Brit. J . Radiol. 57 47-55
Baily N A 1977 Computerized tomography using video techniques Optical
Engineering 16 23-27
-1979 The fluoroscopic image as input data for computed tomography IEEE
Trans. Nucl. Sci. NS26 2707-2709
Baily N A, Keller R A, Jakowatz C V and Kak A C 1976a The capability
of fluoroscopic systems for the production of computerized axial tomograms
Invest. Rad. 11434-439
Baily N A, Lasser E C and Keller R A 1976b Tumour localisation and beam
monitoring-electrofluorotomography Med. Phys. 3 176-1 80
Bamber J 1992 private communication
Bentley R E, Webb S , Leach M 0, Maureemootoo K, Yarnold J R, Toms M A
and Gardiner J 1987 A CT scanner, based on a radiotherapy simulator, as an
aid to planning the conservative treatment of the breast Proc. 24th Kongress
der Gesellschajl Fur Medizinishe Radiologie der DDR Mit Internationaler
Beteiligung (Leipzig) Vol 1 paper V160
Blume S S 1992 Insight and industry: On the dynamics of technological change
in medicine (Camb Mass: MIT Press)
Chen C N and Hoult D I 1989 Biomedical magnetic resonance tomography
(Bristol: Adam Hilger)
Coffey C W, Hines H C, Wang P C and Smith S L 1984 The early applications
and potential usefulness of NMR in radiation therapy treatment planning P roc.

References 329
8th Int. Con$ on the Use of Computers in Radiation Therapy (Toronto, 1984)
(IEEE Computer Society, Toronto) pp 173-180
Cynosure 1985 Cynosure 2000-CTpackage for trans-axial computed tomography
on radiotherapy simulator gantries (Cynosure Imaging Systems Inc, North-
leach, Gloucester, UK)
Dobbs H J and Webb S 1988 Clinical applications of x-ray computed tomography
in radiotherapy planning The Physics of Medical Imaging ed S Webb (Bristol:
Adam Hilger) pp 128-141
Duinker S, Geluk R J and Mulder H 1978 Transaxial analogue tomography Oldelft
Sci. Eng. Quarterly 1 41-66
Dummling K 1970 Einrichtung zur Anfertigung von rontgenologischen Korper-
Querschnittsaufnahmen German patent no 2050825
Feldkamp L A, Davies L C and Kress J W 1984 Practical cone-beam tomography
J . Opt. Soc. Am. A 1 (6) 612-619
Flatman W, Leach M 0 and Webb S 1982 The use of a median window filter with
a simple C T scanner Proc. 6th ICMP (Hamburg, 1982) paper 20: 12
Fraass B A, McShan D L, Diaz R F, Ten Haken R K, Aisen A, Gebarski S , Glazer G
and Lichter A S 1987 Integration of magnetic resonance imaging into radiation
therapy treatment planning: 1. Technical considerationsInt. J.Rad. Oncol. Biol.
Phys. 13 1897-1908
Gifford D 1984 Handbook of physics for radiologists and radiographers
(Chichester: Wiley)
Gilbert P 1972 Iterative methods for three-dimensionalreconstruction of an object
from projections J. Theor. Bid. 36 105-117
Goitein M 1983 Impact and use of CT in planning treatment Advances in Radiation
Therapy Treatment Planning ed A E Wright and A L Boyer (New York:
American Institute of Physics) pp 31Ck320
Guzzardi R 1987 Physics and engineering of medical imaging (Dordrecht:
Martinus Nijhoff) (in collaboration with NATO Scientific Affairs Division)
Hamilton B 1982 (ed) Medical diagnostic imaging systems: technology and
applications (New York: Frost and Sullivan Press)
Harrison R M 1981 Potential applications of computerized tomography to
treatment planning using a radiotherapy simulator Computerized Tomographic
Scanners in Radiotherapy in Europe (Brit. J. Radiol. Supplement 15) chapter
48
Harrison R M and Farmer F T 1976 Possible application of a radiotherapy
simulator for imaging of body cross-sections Brit. J. Radiol. 49 8 13
-1978 The determination of anatomical cross-sections using a radiotherapy
simulator Brit. J. Radiol. 5 1 448-453
Henkelman R M 1992 New imaging technologies: prospects for target definition
Int. J. Rad. Oncol. Biol. Phys. 22 25 1-257
Henkelman R M, Poon P Y and Bronskill M J 1984 Is magnetic resonance
imaging useful for radiation therapy planning? Proc. 8th Int. Conf. on the Use
of Computers in Radiation Therapy (Toronto, 1984) (IEEE Computer Society,

Toronto) pp 181-185
Herman G T 1980 Image reconstruction from projections: The fundamentals of
computed tomography (New York: Academic)
- 1982 Reconstruction algorithms for non-standard CT scanner designs J .
Med. Sys. 6 555-568
Herman G T, Hinds J A, Peretti R W and Rowland S W 1975 SNARK-75-
a programming system for the reconstruction of objects from shadowgraphs
Technical report number 96 (State University of New York at Buffalo, Dept. of
Computer Science)
Herman G T, Lewitt R M, Odhner D and Rowland S W 1989 SNARK-89- a
programming system image reconstruction from projections Report number
MIPGl60 (Medical Image Processing Group, Philadelphia, Pennsylvania)
Hogstrom K R 1983 Implementation of CT treatment planning Advances in
Radiation Therapy Treatment Planning ed A E Wright and A L Boyer (New
York: American Institute of Physics) pp 268-28 1
Hounsfield G N 1968 A method of and apparatus for examination of a body by
radiation such as X or gamma radiation UK Patent no 1283915
- 1973 Computerised transverse axial scanning (tomography): Part 1
Description of system Brit. J. Radiol. 46 1016-1022
Huesman R H, Gullberg G T, Greenberg W L and Budinger T F 1977 RECLBL
Library users manual: Donner algorithms for reconstruction tomography
Lawrence Berkeley Laboratory Publication 2 14
Kak A C, Jakowatz C V, Baily N A and Keller R A 1977 Computerized
tomography using video recorded fluoroscopic images IEEE Trans. on Biomed.
Eng. BME 24 (2) 157-169
Kijewski M F, Judy P F and Svensson G K 1984 Image quality of an analog
radiation therapy simulator-based tomographic scanner Med. Phys. 11 (4) 502-
507
Kotre C J and Harrison R M 1988 Clinical use of simulator-based computed
tomography in Newcastle upon Tyne Brit. J . Radiol. 61 561
Kotre C J, Harrison R M and Ross W M 1984 A simulator-based CT system for
radiotherapy treatment planning Brit. J . Radiol. 57 63 1-635
Kotre C J, Lambert G D and Dawes P J D K 1986 The use of a simulator-based
computed tomography in iridium- 192 dosimetry Brit. J. Radiol. 59 1035-1036
Leach M 0, Flatman W, Webb S , Flower M A and Ott R J 1984a The application
of variable median window filtering to computed tomography Information
processing in medical imaging ed F Deconinck (Dordrecht: Martinus Nijhoff)
pp 151-168
Leach M 0 and Webb S 1979 Application of a radiotherapy simulator for imaging
body cross-sections Proc. HPA Ann. Conf. (Bristol, 1979)
- 1980 A simple CT scanning system suitable for radiotherapy treatment
planning: recent results Brit. J . Radiol. 53 1024-1025
- 1982 A dedicated CT scanner using a position sensitive detector for
radiotherapy treatment planning Proc. 6th ICMP (Hamburg,1982)paper 20: 19

References 33 1
Leach M 0,Webb S and Bentley R E 1982 A rotate translate CT scanner providing

cross-sectional data suitable for planning the dosimetry of radiotherapy
treatment Med. Phys. 9 (2) 269-275
-1985a An x-ray detector system and modified simulator providing CT images
for radiotherapy dosimetry planning Phys. Med. Biol. 30 303-31 1
Leach M 0, Webb S , Bentley R E and Yamold J R 1984b The application of a
radiotherapy simulator-based CT scanner for radiotherapy treatment planning
Phys. Med. Biol. 30 94
Leach M 0, Webb S , Bentley R E, Yarnold J R and Toms M A 1985b Radiotherapy
dosimetry planning using data from a radiotherapy simulator based CT scanner
71st RSNA Con5 (Radiology 157 (P) 3 11)
Ledley R S 1976 Introduction to computed tomography Comp. Biol. Med. 6 239-
246
Ledley R S, Wilson J B, Golab T and Rotolo L S 1974 The ACTA scanner: the
whole body computerized tomograph Comp. Biol.Med. 4 145-155
Lock V and Codes D 1986 Improved localisation for radical treatment of breast
carcinoma using CT and ultrasound Radiography 52 (606) 281-286
Maureemootoo K, Webb S , Leach M 0 and Bentley R E 1988 The performance
characteristics of a simulator-based CT scanner IEEE Trans. Med. Imag. 7 91-
98
McShan D L and Fraass B A 1987 Integration of multi-modality imaging for use
in radiation therapy treatment planning Computer Assisted Radiology ed H U
Lemke, M L Rhodes, C C Jaffee and R Felix (Berlin: Springer) pp 300-304
Moores B M, Parker R P and Pullan B R (eds) 1981 Physical aspects of medical
imaging (Chichester: Wiley)
Natterer S 1986 The mathematics of computerised tomography (New York: Wiley)
New Scientist 1972 X-ray diagnosis peers inside the brain April 27th 1972
Ott R J, Batty V, Bateman T E, Clack R, Flower M A, Leach M 0, Webb S
and McCready V R 1985 Low cost positron computed tomography Recent
developments in medical imaging ed R P Clark and M R Goff (London: Taylor
and Francis) pp 126-1 3 1
Ott R J, Flower M A, Babich J W and Marsden P K 1988 The physics of
radioisotope imaging The physics of medical imaging (Bristol: Adam Hilger)
pp 142-318
Parker R P and Hobday P A 1981 CT scanning in radiotherapy treatment planning:
its strengths and weaknesses Computerised axial tomography in oncology ed J
E Husband and P A Hobday (Edinburgh: Churchill Livingstone) pp 90-100
Phelps M E 1986 Positron emission tomography: principles and quantitation
Positron emission tomography and autoradiography ed M E Phelps, J C
Mazziotta and H R Schelbert (New York: Raven)
Ramachandran G N and Lakshminarayanan A V 1971 Three-dimensional
reconstruction from radiographs and electron micrographs: applications of
convolutions instead of Fourier transforms Proc. Nut1Acad. Sci., USA 68 2236-
2240

Redpath A T 1988a Clinical use of the Edinburgh simulator computed tomography

system Brit. J . Radiol. 61 561
-1988b An analysis of the changes to radiation dose distributions resulting
from the use of simulator computed tomography Brit. J . Radiol. 61 1063-1065
-1988c Proc. HPA Ann. Conk (Brit.J . Radiol. 61 561)
Redpath A T and Wright D H 1985 The use of an image processing system in
radiotherapy simulation Brit. J . Radiol. 58 1081-1089
Schuman W P, Griffin B R, Haynor D R, Johnson J S , Jones D C, Cromwell L D
and Moss A A 1985 MR imaging in radiation therapy planning Radiology 156
143- 147
Smith V and Baker R J 1976b A proposed scheme to implement a transverse
section transmission scanner for radiotherapy treatment planning Computer
applications in radiation oncology ed E S Stemick pp 420423
Smith V, Boyd D P, Kan P T 1977 A computerized transverse section transmission
scanner for radiation therapy treatment planning Phys. Med. Biol. 22 128
Smith V, Boyd D P, Kan P T and Baker R J 1976a A computerised transverse
section transmission scanner for radiation therapy treatment planning Proc. 4th
ICMP Ottawa (Phys. Can. 32 paper 26:6) (Ottawa: Canadian Association of
Physicists)
Smith V, Parker D, Peschmann K, Phillips T, Wang E and Jimbo M 1982 An
improved simulator-based CT scanner for radiation therapy treatment planning
Proc. 6th ICMP (Hamburg, 1982) paper 20: 18
Smith V, Parker D L, Stanley J M, Phillips T J, Boyd D P and Kan P T
1980 Development of a computed tomographic scanner for radiation therapy
treatment planning Radiology 136 489493
Smith V, Stanley J H, Parker D L 1978 Progress towards a CT scanner for radiation
therapy treatment planning Med. Phys. 5 339-340
Sontag M R, Galvin J M, Axel L and Bloch P 1984 The use of NMR images
for radiation therapy treatment planning Proc. 8th Int. Conf on the Use of
Computers in Radiation Therapy (Toronto, 1984) (Toronto: IEEE Computer
Society) pp 168-172
Siisskind C 1981 The invention of computed tomography History of Technology
1981 ed A R Hall and N Smith (London: Mansell Publishing CO)pp 39-80
Swindell W and Webb S 1988 X-ray transmission computed tomography The
physics of medical imaging ed S Webb (Bristol: Adam Hilger) pp 98-127
Taylor J 1988 Imaging in radiotherapy (London: Croom Helm)
Ten Haken R K, Kessler M L, Stem R L, Ellis J H and Niklason L T 1991 Quality
assurance of CT and MRI for radiation therapy treatment planning Quality
assurance in radiotherapy physics ed G Starkschall and J Horton (Madison,
WI: Medical Physics Publishing) pp 73-103
Todd Pokropek A 1989 Theory of tomographic reconstruction Tomographic
methods in nuclear medicine: physical principles, instruments and clinical
applications ed B D Ahluwalia (Boca Raton: Chemical Rubber Company) pp
3-3 3

References 333
Toms M A, Gardiner J, Webb S , Leach M 0, Maureemootoo K and Yamold J R

1986a Clinical dosimetry for breast radiotherapy based on imaging with the
prototype Royal Marsden Hospital CT Simulator Proc. 5th Annual ESTRO
Meeting (Baden-Baden, 1986) pp 207
-1986b Clinical dosimetry for breast radiotherapy based on imaging with the
prototype Royal Marsden Hospital CT Simulator 1st Meeting of the British
Oncological Assoc. (imperial College, 1986)
Varian 1992 Varian Ximatron Simulator/ CT Option Varian Associates technical
literature
Webb S 1982 A modified convolution reconstruction technique for divergent
beams Phys. Med. Biol. 27 419423
- 1987 A review of physical aspects of x-ray transmission computed
tomography Proc. IEE 134 Pt A (2) 126-135
-(ed) 1988 The physics of medical imaging (Bristol: Adam Hilger)
- 1990a From the watching of shadows: the origins of radiological
tomography (Bristol: Adam Hilger)
- 1990b Non-standard CT scanners: their role in radiotherapy int. J. Rad.
Webb S , Flower M A, Ott R J, Leach M 0, Fielding S , Inamdar R, Lowry C
and Broderick M D 1985a A review of studies in the physics of imaging by
single photon emission computed tomography Recent developments in medical
imaging ed R P Clark and M R Goff (London: Taylor and Francis) pp 132-146
Webb S and Leach M 0 1981 Alternative tomographic systems for radiotherapy
planning Computerized axial tomography in oncology ed J Husband and P A
Hobday (Edinburgh: Churchill Livingstone) pp 154-169
Webb S , Leach M 0, Bentley R E, Maureemootoo K, Nahum A, Toms M A,

Webb S , Leach M 0 and Newbery P 1988 CT scanner and detector therefor UK

Patent no 2 170980, US Patent no 4769829, European Patent no 0194743
Webb S , Leach M 0 and Yarnold J R 1984c CT of the breast achieved in the
treatment position using a CT facility developed for a radiotherapy simulator
Proc. 3rd Annual ESTRO Meeting (Jerusalem, 1984) p 215
Webb S , Lillicrap S C, Steere H and Speller R D 1977a Application of a
radiotherapy simulator for imaging of body cross-sections Brit. J. Radiol. 50
152- 153
- 1977b Application of a radiotherapy simulator for imaging body cross-
sections Brit. J. Radiol. 5 1 158
Webb S , Lillicrap S C, Steere H, Speller R D and Jones J 1978 Reconstruction of
body cross-sections for use in radiotherapy treatment planning Computers in
Radiation Therapy ed U Rosenow pp 749-753
Webb S and Long 1985 The effect of reversing projection data in reconstruction
tomography Phys. Med. B i d . 30 81-83
Webb S , Speller R D and Leach M 0 1981 The generation of CT data
for radiotherapy planning using devices other than commercial scanners
Computerized Tomographic Scanners in Radiotherapy in Europe (Brit. J .
Radiol. Supplement 15) pp 178-181
Webb S , Sutcliffe J, Burkinshaw L and Horsman A 1987c Tomographic
reconstruction from experimentally obtained cone-beam projections IEEE
Trans. Biomed. Imag. MI6 (1) 67-73
Wells P N T (ed) 1982 Scientific basis of medical imaging (Edinburgh: Churchill
Livingstone)
Wright D H, Redpath A T, Jarvis J G H and Hams J R 1984 Image processing
techniques applied to fluoroscopic x-ray pictures obtained from a radiotherapy
simulator Proc. 8th Int. Conf on the Use of Computers in Radiation Therapy
(Toronto, 1984) (IEEE Computer Society, Toronto) pp 343-349

EPILOGUE
The goal of conformal radiotherapy is clearly not new, but the technology allowing
progress towards achieving this goal is now under serious active development
and the subject has a rejuvenated topicality. Physicists and engineers are
now able to manufacture equipment for the practical and accurate delivery of
radiation, and computer science has matured to the point that systems for planning
radiotherapy, controlling radiation delivery and verifying its positional accuracy
are available. Medical imaging ‘drives’ conformal radiotherapy planning and
recent improvements in 3D imaging have led naturally into improvements in
conformal radiotherapy. As might be expected, progress has inevitably spawned
new questions. For example, the prediction of biological effect is still in its
infancy by comparison. Ironically there is also concem that too precise a definition
of target volume might not include microscopic peripheral spread of malignant
disease. This serves to emphasize that the physicist is not the person to solve
all the problems. There is a need for close collaboration with radiobiologists,
diagnostic radiologists, radiotherapists and other cancer medicine clinicians.
Achieving conformal radiotherapy relies on optimizing a chain of events; any
weakness in one element of the chain weakens the whole.
At the heart of planning conformal radiotherapy are computer systems, able to
accept, register, display and analyse 3D medical images from different modalities.
In the 1980’s these were largely to be found as ‘one-off’ developments in
university hospitals, but the 1990’s have seen a number of systems become
available commercially. They are not cheap, but one may expect the cost to fall
even as their features expand.
Techniques to optimize dose are actively under development. Some are
realizable in practical computational times; others giving ‘more conformal’ 3D
dose distributions take longer and some invoke technology which is not readily
available, such as intensity-modulated fields.
The availability of the multileaf collimator opens up new and exciting
possibilities for treatment delivery and should be able to make the repetitive
casting of shielding blocks redundant. Under computer control it creates new
opportunities for dynamic field shaping which are quite unachievable with block
casting. It may even solve the problem of delivering fields with spatially varying
beam intensities. Dynamic therapy with moving gantry, couch and collimator
jaws will undoubtedly continue, but it will be interesting to see to what extent the
335
336 Epilogue
multileaf collimator takes over as the preferred method of tailoring the geometry
of the fields.
Stereotactic radiation of brain disorders, malignant and benign, will continue
apace with linear accelerators. Enough work has now been done to demonstrate
that with just a few arcs, very satisfactory dose distributions can be obtained. It
will be interesting to see whether the gamma knife continues to compete. Perhaps
the problems will divide into two classes appropriate to each radiation treatment
machine?
Electronic portal imaging is the key to quantifying patient registration between
imaging, treatment planning and the time of treatment delivery. One may expect
that soon it will become routine to check each field and, if necessary, reposition the
patient. Several studies are addressing this problem and more will undoubtedly
do so.
The use of protons continues to be rare, despite their obviously superiorphysical
properties. The reason is largely one of cost. The first purpose-built, as opposed to
shared, proton treatment facility has just come on stream and the community looks
to evaluate its success for the justification of future proton irradiation facilities.
The application of the physical sciences to medicine has undoubtedly been a
major contribution to radiotherapy and will continue to be so, as far as developing
technology is concerned. Much has been achieved, but much remains to be
done and this book has attempted to cover the major areas of activity. It cannot,
however, cover all the problems and, in particular, it has not attempted to be too
definitive where precise knowledge is not to hand. In particular, the translation of
dose to biological effect is currently based on less data than one would like, and
a developing area will undoubtedly be the collaboration of physicists and clinical
partners to put right this deficiency. Finally, all sponsors expect a retum on their
investments and there are still unanswered questions conceming the efficacy of
conformal radiotherapy and some doubts about the best moment in the course of
research to mount the necessary exercises in assessment.

APPENDIX A
NUMERICAL QUESTIONS
The following numerical questions may be helpful when teaching from this book.
To answer the questions will require data which can be found in tables of physical
constants such as Hubbell (1969,1982), Hubbell et a1 (1980), Duck (1990) and
Kaye and Laby (1986). All the relevant theory is covered in this book. Some
calculations require writing a short piece of computer code.
Chapter 1
1. The tumour control probability (TCP) for a tissue is specified by an a value
from a Gaussian distribution of mean 0.35 and standard deviation 0.08. The
clonogenic cell density is 1O7 per cc. If a target volume of 160 cc is irradiated
to a uniform dose of 60 Gy, what is the predicted TCP with the cell-kill model?
What is the TCP for the same volume at a dose of 65 Gy? What is the TCP for
a larger volume of 320 cc at the first dose of 60 Gy? (You may need to write a
computer program to do this question.)
2. The normal tissue complication probability (NTCP) for a uniform irradiation
of the full volume of an organ at risk is 0.2 (i.e. 20%). What would be the NTCP
if only half the organ were raised to the same dose? What would be the NTCP if
instead 0.1 of the organ were raised to the same dose?
3. The full volume of an organ at risk is raised to 60 Gy, giving a normal
tissue complication probability of 0.1 (10%). What would be the equivalent
dose delivered to half the organ to give the same normal tissue complication
probability? (Assume k = 10 for this organ.)
4. A volume containing an organ at risk in a radiotherapy treatment has a (0.5)
50% normal tissue complication probability when irradiated uniformly to a dose
of 50 Gy. What would be the NTCP if the dose were (i) 40 Gy (ii) 30 Gy? (Assume
k = 10 for this organ.)
5 . A cumulative dose-volume histogram is specified by the formula
V ( d )= 1 -
{ 1
1
+ [(121 - d ) /41]" I
337
338 Numerical questions
where d is the dose in Gy (in the range 1 to 120 Gy) and V ( d )is the volume raised
to at least this dose. Plot a graph of V ( d ) versus d . What is the effective fractional
volume raised to the maximum dose of 120 Gy according to the Kutcher-Burman
reduction method? (Take n = 0.1 and you may need to write a computer program
to do this question.)
6 . Two imaging modalities specified on a rectangular Cartesian coordinate
system are related as follows:
0 a rotation by 90" about the x axis,
0 a rotation by 90" about the y axis,
0 no rotation about the z axis,
0 an isotropic scaling by a factor 0.5,
0 a shift by 1 cm in the x direction, by 2 cm in the y direction and by 3 cm in the
z direction.
Write down the coordinate transformation between the two imaging modalities
as a matrix multiplication only.
Chapter 2
7. A radially symmetric dose distribution (see equation (2.9)) is given by
rotating a non-uniform beam profile specified by a cosine dependence along a
radius through the origin of rotation coordinates up to a radius of 5 cm and
zero thereafter (specified by equation (2.22)). Assume that ~ = 0 . 0 2cm-' for
the radiation. Write a small piece of computer code to plot a graph of the dose
distribution as a function of radius up to a distance of 10 cm. Discuss how the
dose distribution falls off for radial distances beyond 5 cm. Why is the dose not
zero in this region?
8. Use the code from question 7 to vary the radius beyond which the beam
profile is zero inwards from 5 cm to (sequentially) 3 cm, 1 cm, 0.1 cm. What do
you notice about the dose distribution up to a radial distance of 10 cm as this is
done? What is the name given to the last of these distributions? How is it used in
'convolution dosimetry'?
9. A rotationally symmetric dose distribution is generated in which the dose is
zero inside a radius of 4 cm and uniform from 4 to 8 cm. Plot the intensity profile
at the source in the region x lying between 4 and 8 cm required for radiation with
F = 0.02 cm-'. Assume the phantom is a circle of radius 15 cm with its centre at
the axis of rotation.
10. A rotating beam of the type shown in figure 2.19 can be considered to deliver
a dose of 0.1 arbitrary units under the central portion between a distance of f1 cm
and 1 unit under the region between f(1-6) cm. Plot the radial distribution of the
circularly-symmetric dose distribution for radii 0 to 10 cm.
11. What is the advantage in terms of speed for computing a 3D convolution
of two 204S3 matrices in Fourier space compared with real space? What is the
corresponding advantage for a 2D convolution of two 512' matrices?

Numerical questions 339
Chapter 4
12. A narrow beam of protons diverges to irradiate a circular area of radius 4
cm at a dose-measurement plane. A simple beam-stopper is placed on axis such
that its projection on the dose-measurement plane fills a circle of radius A 1 = 1
cm. Plot the distribution of proton intensity in the measurement plane if the proton
beam has been scattered twice, firstly by a foil with RI = 1.7A 1 and secondly by
a foil with RZ = 1.4A I just behind the beam-stopper. (You will probably need to
write a short computer code for this question.)
13. A narrow beam of protons diverges to irradiate a circular area of radius
4 cm at a dose-measurement plane. An annular beam-stopper is placed on axis
such that its projection on the dose-measurement plane fills an annulus of outer
radius A1 = 1 cm and inner radius A2 = 0.38 cm. Plot the distribution of proton
intensity in the measurement plane if the proton beam has been scattered twice,
firstly by a foil with RI = 1.7A1 and secondly by a foil with R2 = 1.OA1 just
behind the beam stopper. (You will probably need to write a short computer code
for this question.)
14. Given that a 100 Mev proton has a range in tissue of about 8 cm, what
would be the approximate energies of an alpha particle and a deuteron with the
same range?
15. If a proton beam facility can produce A cm-2, what dose rate would
be delivered to the last 1 cm of the proton range?
chapter 5
16. A multileaf collimator comprises 40 pairs of leaves. Each leaf is 5 mm wide.
The leaves are 40 cm from the x-ray target, can open to a maximum of 10 cm on
the leaf-side of mid-line and can over-run the mid-line in either direction by 6 cm.
What is the largest field that can be set at the isocentric distance of 1 m?
17. A multileaf collimator is made of tungsten leaves 5 cm thick. What
fraction of 2 MeVmonoenergetic photons ‘leaks’ through any closed portion of the
collimator (ignoring the possible hairline gap between any two adjacent leaves)?
If in addition a back-up collimator 2 cm thick resides behind the leaves, to what
fraction is the leakage reduced at this energy?
18. A multileaf collimator is made of tungsten leaves 5 cm thick and they are
‘stepped’at a depth of 2.5 cm along the edges of the leaves. What leakage radiation
could be expected at the step for monoenergetic 2 MeV photons if there were no
back-up collimator?
chapter 6
19. Monoenergetic x-rays at 2 MeV are incident normally on the surface of
a 25 mm thick slab of the scintillator bismuth germanate. Calculate the fraction

340 Numerical questions
of the x-ray flux which is stopped by the scintillator. What fraction of the flux
would be stopped if the scintillator were zinc tungstate instead? Calculate the
same fraction for these two materials if the x-ray energy were increased to 6 MeV.
How thick should a scintillator of the material caesium iodide be to stop 75%
of x-ray photons of energy 1 MeV? What are the implications for megavoltage
imaging?
20. The detector for megavoltage imaging comprises a 1 mm thick plate of
copper bonded to a 300 mg cm-’ thick scintillator of gadolinium oxysulphide.
Calculate the fraction of 1 MeV monoenergetic x-rays which stop in the copper
plate. What would be the consequences, of making the copper plate thicker, for
detective quantum efficiency and spatial resolution?
21. A fluoroscopic megavoltage imaging system images the light from a
scintillator of area 400 mm2 on to a CCD camera with a face size of 4 mm2 via
an optical lens with F-number 0.9 and optical transmission coefficient 0.8. If the
photon detection efficiency of the CCD camera is 70% and 2 x IO4 optical photons
are emitted by the scintillator for each detected x-ray photon and the detective
quantum efficiency of the metal-plate converter is 3%, assuming unity refractive
index for the scintillator, by how much is the signal-to-noise ratio of the input
x-ray flux degraded by the input stage of the CCD camera?
22. A digital portal-imaging system based on a fluoroscopic detector creates
images with an improved signal-to-noise ratio by frame averaging. By how much
does the signal-to-noise ratio improve if (i) 16, (ii) 256 frames are averaged? What
are the consequences for static and dynamic imaging?
23. In a fluoroscopic megavoltage imaging system, should the F-number of the
optical component of the system be as small as possible or as large as possible,
and why?
24. In a megavoltage imaging system comprising a scanning array of
photodiodes, the photodiodes are encapsulated in lead 1.5 mm thick. Estimate
the fraction of monoenergetic x-rays of energy 4 MeV which are detected.
25. What is the dose contrast on a film introduced by an extra thickness of 2
mm of water-equivalent tissue for 1 MeV monoenergetic radiation if the scatter-
to-primary ratio is 30%? What is the film density contrast if the y of the film is
2.7 in the region where these intensities occur?
26. Contrast improvement in portal imaging is to be achieved by ‘gamma
multiplication’ using three films. If the first film has a gamma of 2.5, the third
has a gamma of 3.2, what should be the gamma of the second (middle) film if the
overall gamma multiplication is to be 22.5?
27. An image intensifier is used for capturing an image of a patient being
positioned for radiotherapy on a simulator for a square field whose side is 20 cm
long. If the displacement S of an image point from its true position is given by the

References 341
formula
S=aR2+BR4
where R is the true distance of that point from the origin, what is the maximum
displacement which a point suffers if a = 6 x pixels-' and B = 4 x
lo-* pixels-3 and the pixel size is 2 mm?
REFERENCES
Duck F A 1990 Physical properties of tissue: A comprehensive reference book

(London: Academic)
Hubbell J H 1969 Photon cross sections, attenuation coefficients and energy
absorption coefficients from 10 keV to 100 GeV Report number NSRDS-NBS
29, US Govemment Printing Office, Washington DC
Hubbell J H 1982 Photon mass attenuation and energy absorption coefficients from
1 keV to 20 MeV Int. J . Appl. Rad. Isot. 33 1269-1290
Hubbell J H, Gimm H A and 0verbpr 1980 Pair, triplet and total atomic cross
sections (and mass attenuation coefficients) for 1 MeV-100 GeV photons in
elements Z = 1 to 100 J . Phys. Chem. Ret Data 9 (4) 1023-1147
Kaye G W C and Laby T H 1986 Tables of physical and chemical constants
(London: Longmans) 15th edn

APPENDIX B
GLOSSARY OF TERMS
This glossary also includes some widely used terms in medical imaging, which is
so important to conformal radiotherapy.
ANGER CAMERA Gamma-ray detector, named after its inventor, in which
a large-area scintillation crystal fitted with a collimator is viewed by a matrix of
photomultiplier tubes with electronics to give position sensitivity.
ARTEFACT An unwanted structure in an image generally due to some
unsatisfactory (but often insurmountable) feature in imaging equipment.
AUTOTRACKING Method of obtaining the contour of a specific tissue from
images automatically by giving the computer features of the tissue which enable
it to be differentiated from its surroundings (e.g. its linear attenuation coefficient).
BANDLIMIT The maximum frequency present in an image. Real information
at frequencies above the limit fail to be recorded.
BEAM PROFILE The shape of the intensity distribution in a radiation field
having an intensity varying in one or two dimensions.
BEAM SPACE Mathematical space in which beams are defined, possibly with
spatially varying intensities (see also DOSE SPACE).
BEAM-STOPPER Material placed in a charged-particle beam and which
completely absorbs the beam. A combination of a beam-stopper with an annular
shape and a double-scattering system can create a large-area field for radiotherapy.
BEAM’S-EYE-VIEW The view of the patient anatomy as seen through the
radiation collimator by an imaginery observer at the source location. The view
is generally constructed from tomographic image data. It is particularly helpful
when planning to avoid irradiating an organ-at-risk.
BEAM-WEIGHT Rather loose term applied to radiation fields delivered by a
linear accelerator, meaning the number of monitor units per field or part of a field
if this has a spatial variation.
BIOLOGICAL RESPONSE The response of normal and pathological tissues
to radiation. Calculating this requires a model for the damage produced by a
known radiation dose.
BLOCKED FIELDS Radiation fields defined by a collimator to which
additional lead blocks have been added to create a non-rectangular-shaped field.
342
Glossary of terms 343
BLOCKING TRAY Tray in which blocks are secured at the linear accelerator
during photon treatment.
BODY SCANNER Rather loose term usually refemng to a machine for
computed tomography of the body (rather than the head).
BOLUS Material placed next to the patient to change the contour as it appears
to the radiation beam.
BRAGG PEAK The region of high-energy deposition at the end of the track
of a particle (e.g. a proton) slowing down in a tissue.
BUILD-UP REGION A volume in a patient where the electrons from photon
interactions are not in equilibrium. This may be near the patient surface, or an
interface between two materials with widely varying electron densities.
CCD CAMERA Television camera using a solid-state charged-coupleddevice.
These are found in some megavoltage imaging systems and also in mechanisms
for verifying the leaf positions of a multileaf collimator.
CLASSICAL TOMOGRAPHY Generic term for all non-computed or so-
called ‘blurring tomography’
CLINICAL TRIAL Comparison of the effectiveness (with some clearly
specified criteria) of a new method of radiotherapy against some standard and
commonly accepted technique. Patients should be randomized into two arms of
the trial, one set of patients being treated with the ‘old’ method and the other
set receiving the new technique. Ideally neither the patients nor their clinicians
should know which patients are in which arm until the trial is analysed (double
blind trial).
CLONOGENIC CELLS Tumour cells which are reproducing.
COINCIDENCE DETECTION Form of imaging wherein two gamma rays
are detected simultaneously (or nearly so). The gammas may be those from a
positron-emitting radionuclide or an isotope emitting two gammas in cascade.
COLOUR WASH Method of shading grey-scale anatomical tomograms with
colours to indicate the value of dose at each region.
COMPENSATOR Material placed in the path of radiation before it enters a
patient, designed to compensate for the effect of tissue inhomogeneity in the path
of the radiation. Compensators are constructed on different principles for photons
and charged particles.
COMPUTED TOMOGRAPHY (CT) Section imaging in which the required
image must be reconstructed from projection measurements, usually using a
digital computer.
CONFORMAL THERAPY In the old Japanese definition, term for rotation
radiotherapy; in the modem definition, term for radiotherapy with the high-dose
volume geometrically tailored to the target.
CONVOLUTION AND BACKPROJECTION Mathematical technique for
reconstructing CT data from (filtered, or convolved) projections.
CONVOLUTION DOSIMETRY Method of computing dose distributions
based on a knowledge of the elemental dose distribution from radiation
interactions at a point.

344 Glossary of terms
COPLANAR FIELDS Radiation fields with central axes all lying in a common
plane.
CORONAL SLICE Tomographic section bordered by superior, inferior, right
and left of the patient (i.e. face-on view); generally not obtained by direct
reconstruction, but by reorganizing transaxial sections.
COST FUNCTION Mathematical function parametrizing the effect of
arranging beams in some particular way. For example, a simple cost function
could be the RMS difference between the prescribed dose and the delivered
dose. More complicated functions could include biological models. The aim
of optimization would be to minimize the cost function, possibly subject to
constraints.
COUCH TWIST When the axis of rotation of a couch coincides with the axis
of rotation of a radiotherapy gantry, a radiation field which does not have its axis
in the transaxial plane of the patient may be achieved by rotating the couch so
it is not perpendicular to the plane of rotation of the gantry. This is colloquially
referred to as couch twist. It allows non-coplanar beams to be arranged.
CROSSTALK Information leaking from one place to another (generally
adjacent) place; for example, when an x-ray impinges on one element of a matrix
of radiation detectors, another element may record a smaller but non-zero signal.
CYCLOTRON Circular particle accelerator which may be used for the
production of certain radionuclides and for charged particles for radiotherapy.
DECONVOLUTION Inverse process to convolution. Deconvolution of a dose
kemel from a dose distribution can yield the irradiation density for example.
DIGITAL RADIOLOGY Form of radiology (tomographic or non-tomographic)
in which the image is captured (often by discrete detectors) and stored as a digital
matrix of numbers.
DIGITAL RECONSTRUCTED RADIOGRAPH Planar radiograph con-
structed by ray-tracing from the position of an x-ray source through the 3D matrix
of CT data. This image can be compared with a portal image at the time of treat-
ment or with the image from a simulator of the patient at treatment simulation.
DOSE-DIFFERENCE PLOT Map of the difference between the desired dose
distribution (the dose prescription) and the dose distribution resulting from some
particular arrangement of beams.
DOSE KERNEL The distribution of dose around a point at which a photon has
interacted, computed from an ensemble of all the possible interactions.
DOSE SPACE Mathematical space in which dose is defined. The radiation
planning process relates dose space to beam space.
DOSE-VOLUME HISTOGRAM (DVH) Histogram showing the fraction of
the target volume raised to a particular dose value (differential DVH) or the fraction
of the volume receiving the specified dose value or greater (integral DVH).
DOUBLE FOCUSING Arrangement by which the leaves of a multileaf
collimator are engineered. The tangents to all sides and ends of leaves converge
to the source of radiation. This limits penumbra.
DOUBLE-SCATTERING SYSTEM Method by which a narrow proton beam

is broadened to a wide area useful for radiotherapy. The technique uses two thin
metal foils for scattering.
DYNAMIC THERAPY Form of radiotherapy with the beam moving relative
to the patient during the irradiation (opposite of STOP-AND-SHOOT). Sometimes
this is engineered by moving the source. Sometimes, instead, the couch is
translated relative to the source.
DYNAMIC WEDGE The creation of a wedged field by moving one jaw of an
accelerator collimator during irradiation.
EGS4 The electron gamma shower computer package much used for radiation
dose computations in benchmark situations.
ELECTRONIC PORTAL IMAGING Recording the exiting radiation at the
time of treatment with an electronic method without the use of film (sometimes
also called megavoltage portal imaging (MVI)).
EMISSION TOMOGRAPHY Body-section imaging of the distribution of
uptake of a radiopharmaceutical or some other property of its metabolism.
EMISSION COMPUTED TOMOGRAPHY (ECT) Emission tomography
requiring the use of a digital computer to reconstruct digital tomograms. When
based on the detection of single photons, the technique is known as SINGLE-
PHOTON EMISSION COMPUTED TOMOGRAPHY (SPECT). When based on
the coincidence detection of high-energy photons from positron emission, the
technique is known as POSITRON EMISSION TOMOGRAPHY (PET).
ERROR PLAN A dose plan in which the effect of certain known or estimable
sources of error (such as tissue movement) is taken into account. (Not a plan with
errors!)
FIDUCIAL A marker fixed to the patient which will show as a point in a
medical image. Multiple fiducials play an important part in image registration.
FIELD OF VIEW Region in space which is viewed at all times during
tomography.
FILM Generally here meaning film sensitive to x-rays, not visible light.
FILM GAMMA The slope of the linear part of the curve expressing film
density as a function of dose.
FILTERED BACKPROJECTION The technique (or the image created
thereby) by which CT data are reconstructed from filtered projections.
FIRST-GENERATION CT Form of rotate-translate CT using a single
detector (see ROTATE-TRANSLATE CT).
FLUENCE Photons per unit area.
FLUOROSCOPIC SCREEN Detector which emits light photons when x-ray
photons are incident upon it.
FOURIER TRANSFORM Mathematical transformation, named after its
inventor, which converts data from a function of some variable (e.g. time) into
the corresponding function of the inverse variable (e.g. frequency).
FOUR VECTOR Vector with four components. Rotations, scaling and
translations can all be handled by matrix multiplications if points in space are
represented as four vectors.

FOURTH-GENERATION CT Form of rotate-only CT in which the detectors

comprise a stationary ring and the x-ray source only rotates.
FRACTIONATION Process of delivering radiation in a series of small doses
spaced out in time.
FRAME AVERAGING Finding the mean image of a number of (generally
sequential) video frames to reduce the effect of noise.
FRAME GRABBER Computer-based device for creating a digital image from
a video image.
GAMMA CAMERA Usually means Anger camera (see above).
GAMMA KNIFE Specialized irradiation device designed by Leksell for
stereotactic radiosurgery. It contains a large number of cobalt sources focused
to a point.
GAMMA MULTIPLICATION The method of increasing the contrast on a
film by making sequential copies on to films with a film gamma greater than one.
GRAVITY-ASSISTED BLOCKING Method by which a block is suspended
in the field from a rotating treatment machine. The block shields specified volumes
of tissue, whatever the orientation of the gantry.
HAT-HEAD METHOD Technique for registering two image data-sets of the
head. Surfaces are constructed from each and fitted like a hat to a head to minimize
the space between.
HEAD SCANNER Rather loose term, usually referring to a machine for
computed tomography of the head (rather than the body).
HEAD TWIST Colloquial term for rotating the collimator of a treatment
machine so the sides of a square field would not lie parallel to, and orthogonal
to, the plane of rotation of the gantry.
HELMET Collimator for the gamma knife.
IMAGE OF REGRET A colour-washed grey-scale image of anatomy,
wherein the colour codes the inability of a technique to produce a dose distribution
matching the prescription (regret).
IMAGE INTENSIFIER Optoelectronic device creating an optical image from
an x-ray flux impinging upon it.
INDEPENDENT COLLIMATOR X-ray collimator on a linear accelerator
with the ability for each jaw to move independent of its paired jaw.
INTENSITY MODULATION FUNCTION Mathematical function describ-
ing the variation of radiation intensity in beam space.
INVERSE COMPUTED TOMOGRAPHY Somewhat loose term meaning
the process of computing the optimum distribution of beam intensities which will
combine to yield some specified dose distribution. When formulated in two-
dimensions, the problem has certain similarities with the mirror of the method
of reconstructing CT slices from 1D projection data, hence the name.
INVERSE RADON TRANSFORM The mathematics embodied in recon-
struction theory for computed tomography.
IRREGULAR FIELD Rather loose term, generally meaning radiation field
which is not rectangular and which can therefore not be collimated by a pair of

simple jaws without the addition of blocks.

ISOCENTRE The axis of rotation of a gantry, for example that of a CT scanner
or a machine delivering radiotherapeutic x-radiation.
ISODOSE LINE/SURFACE Line (in a 2D plan) or surface (in a 3D plan)
connecting points where the dose is the same.
ISOEFFECT FORMULA Equation giving the relationships between quanti-
ties such as time, dose per fraction and number of fractions, which may be varied
to give the same biological effect.
KERMA Kinetic energy released per unit mass. Kerma includes energy
released by charged particles only. It excludes radiation transport from scattered
photons. Kerma has the same dimensions as dose (compare with TERMA).
KICKER Device for extracting a charged particle beam from an accelerator.
LANDMARK-BASED CORRELATION Method of registering two (gener-
ally 3D) image data-sets by identifying common (generally internal anatomical)
landmarks in the two images.
LARGE-SCREEN DIGITIZER Device for projecting an image of a patient
cross section and some means of pointing at the image. By tracing a feature of
interest, the required contour is input to the computer.
LEAKAGE RADIATION Unwanted radiation for example passing through
the gaps between closed collimators or adjacent leaves of a multileaf collimator.
LOCAL CONTROL Control of the primary tumour without metastatic spread.
MATCH-LINE PROBLEMS Potential sites of overdose or underdose where
two adjacent radiation fields meet.
MATRIX IONIZATION CHAMBER A type of megavoltage portal imager.
MEGAVOLTAGE COMPUTED TOMOGRAPHY (MVCT) CT performed
on a treatment machine at the time of treatment, used to verify the patient
positioning.
MEGAVOLTAGE PORTAL IMAGING (MVI) Making a picture of the
radiation emerging from the patient at the time of treatment.
METAL SCREEN Thin sheet of metal placed in front of a film or scintillator
to convert an x-ray flux into a detectable electron flux.
MODALITIES Techniques whereby medical images may be obtained by
physical probes.
MONTE CARLO METHODS Calculation techniques using a computer,
whereby the macroscopic effect of an irradiation is constructed as the ensemble
average of the individual effects of following the scattering history of discrete
photons or particles. Millions of histories need to be computed for accurate results.
The accuracy of the results is limited only by the input data, the sophistication of
the modelling of the interactions and the amount of available computer time.
MORBIDITY Treatment-related complications which adversely affect the
quality of life of a patient.
MULTICRYSTAL PET SCANNER Form of positron tomography device
using many scintillation crystals usually arranged in a circle.
MULTILEAF COLLIMATOR X-ray collimator comprising many ‘leaves’

and thus able to create an irregularly shaped window for the radiation to pass
through. Typically there might be some 40 pairs of these leaves with each leaf
projecting to a width of about 1 cm at the isocentre.
MULTIPLE-BEAM RADIOTHERAPY Radiotherapy of tumours by com-
bining one or more x-ray beams from different directions relative to the patient.
MULTIPLE SCATTERING The processes by which photons or charged
particles lose energy in traversing tissue. Photons may only scatter a few times
before being annihilated, whereas electrons and protons undergo thousands of
small scattering events.
MULTIWIRE PROPORTIONAL CHAMBER (MWPC) Form of gamma-
ray detector based on proportional chamber theory.
NORMAL TISSUE COMPLICATION PROBABILITY (NTCP) Graph of
the probability of inducing a complication in a normal healthy tissue as a function
of the applied radiation dose. It is necessary to specify the ‘end-point’, the specific
complication appropriate to the curve (e.g. rectal bleeding).
OBLIQUE TOMOGRAPHY Form of tomography wherein arrangements are
made for the slice in sharp focus to lie in some plane other than transaxial, coronal
or sagittal.
OPTIMIZATION Technique to establish the best arrangement of radiation
fields to maximize uncomplicated tumour control. The word is widely used, with
varying shades of interpretation.
ORGAN-AT-RISK Sensitive structure requiring shielding during radiotherapy
(e.g. the spinal cord).
OVER-RUNNING Ability for the leaves of a multileaf collimator to be set past
the axis of rotation of the treatment machine, i.e. the beam port is entirely to one
side of the axis of rotation.
PARALLEL-BEAM CT Form of computed tomography in which all the ray-
paths to projection elements (at each orientation) are parallel (i.e. 1st generation
CT).
PARALLEL-HOLE COLLIMATOR Collimator for an Anger camera, also
enabling SPECT in parallel (as opposed to fan- or cone-) beam geometry.
PARALLEL-OPPOSED FIELDS Radiation fields sharing a common axis and
pointing towards each other.
PARTIAL VOLUME EFFECT The effect whereby the value of a physical
parameter ascribed to a voxel is some average value due to the voxel spanning
tissues of different types.
PENCIL-BEAM DOSE DISTRIBUTION The distribution of dose surround-
ing a narrow parallel beam of radiation with an infinitessimal cross section. This
forms the elemental kemel for some methods of dose calculation.
PENUMBRA In radiology, term meaning the part shadows caused by (among
other things) finite source size and collimation. In radiotherapy, term meaning the
region at the edge of a field where the dose rapidy decreases, but does not become
zero due to scatter, as well as the above.
PHANTOM Object generally comprising tissue substitute materials used to

Glosswy of terms 349
simulate a patient or part thereof.

PHOTOSTIMULABLE PHOSPHOR PLATE Radiation detector able to
store an image for later digital readout.
PIN-CUSHION DISTORTION Geometrical distortion, symmetrical about
the detector centre, misplacing the detected event on an image intensifier.
PIXEL Picture element in a digital image matrix (for example, a CT S C ~ ) .
POINT IRRADIATION DISTRIBUTION Distribution of dose at a point
resulting from the superposition of many narrow radiation fields converging on
that point, for example by rotation therapy.
POINT-SPREAD FUNCTION The response of an imaging system to a point
source of radiation.
PORT The opening, usually defined by a collimator, through which radiation
emerges to impinge on a patient. On the distal side of the patient the emerging
radiation can be recorded by ‘portal imaging’.
POSITION-SENSITIVE DETECTOR A radiation detector in which the
position of the x-ray interaction can be very precisely determined. Some position-
sensitive detectors are multi-element (e.g. crystal scintillators); other continuous
detectors use position-sensing electronics (e.g. as in the Anger or gamma camera).
POSITRON EMISSION TOMOGRAPHY (PET) Form of sectional imaging
of a radiopharmaceutical, in which the radionuclide is a positron emitter (in
contrast to a single-gamma emitter).
PROJECTION The sum along a particular (usually straight) line path through
a patient of some physical property of biological tissue. In transmission computed
tomography the projections are sums of x-ray linear attenuation coefficient. In
emission computed tomography the sums are complicated functions of both
the uptake of a radiopharmaceutical in body tissues and the photon attenuation
properties of the biological tissue.
PROJECTION FILTERING A vital stage in the mathematical process of
reconstructing cross-sectional images from projections (CT). After projection
filtering, data are back-projected into the reconstruction space.
PROTON CT Form of computed tomography using protons as probe.
QUANTUM EFFICIENCY Fraction of photons which impinge on a detector
which contribute to the image (for example, in megavoltage imaging).
RADIODENSITY Loose term for x-ray linear attenuation coefficient (which
is related to the electron density of tissue).
RADIONUCLIDE Isotope of an element which is radioactive.
RADIOPHARMACEUTICAL Pharmaceutical, element or compound la-
belled with a radionuclide.
RADIOTHERAPY TREATMENT FRACTIONS The delivery of part of a
radiotherapy treatment via a series of visits for the patient.
RANGE The maximum distance a charged particle will travel in a tissue.
RANGE MODULATION The process of reducing the range in a patient of a
charged-particle beam by placing absorbing material in its path upstream of the
patient.

RANGE-MODULATING WHEEL OR PROPELLOR Rotating wheel

comprising a fan of modulators, each of different thicknesses, which, when placed
in a monoenergetic proton beam, generates a beam with a mixture of ranges. The
carefully calculated geometry of the wheel creates an approximately flat spread-
out Bragg peak.
RECONSTRUCTION ALGORITHM The mathematical method by which
an image may be created from measurements of some physical property of
biological tissues (such as x-ray linear attenuation coefficient or the uptake of a
radiopharmaceutical) along a large number of different paths through the patient.
REDUCTION SCHEME method of converting the dose-volume histogram
of a volume which receives an inhomogeneous radiation dose to either ( a ) an
equivalent volume receiving a homogeneous single dose or ( b ) an equivalent
single dose to the whole volume.
REGISTRATION Process of aligning two (generally 3D)image data-sets so
that a region in one exactly corresponds to the same region in the patient in the
other.
REGRET Inability to meet a dose specification (see VOLUME OF REGRET
and IMAGE OF REGRET).
RISK CURVE Loose term for TCP or NTCP curve.
ROTATE-ONLY CT Form of computed tomography in which an x-ray source
delivering a fan-beam of radiation on to a bank of detectors (the fan completely
spanning the patient) rotates (with no translation) about the patient: if the source
and detectors both rotate then this is also known as 3rd generation CT scanning. If
only the source rotates and the detectors comprise a stationary ring, this is known
as 4th generation CT.
ROTATE-TRANSLATE CT Form of computed tomography in which a single
x-ray source and single detector (or small bank of detectors) are translated in a
straight line relative to a patient, with subsequent rotation of the line of translation
through a large number of orientations relative to the patient. The PROJECTIONS
so recorded form the basis of reconstruction. The form of CT with a single
detector is also known as 1st generation CT; the form with a bank is known as
2nd generation CT.
SAGITTAL SLICE Tomographic section bordered by superior, inferior,
anterior and posterior of the patient (i.e. side-on view); generally not obtained
by direct reconstruction but by re-organizing transaxial sections.
SCATTER Radiation which results from the interaction of primary radiation,
for example by the Compton effect in tissue.
SCINTILLATION DETECTOR Form of x-ray detector in which x-ray
energy is converted into optical photons.
SCOUT VIEW A digital two-dimensional planar radiograph formed using a CT
scanner, operated such that the source and detector do not rotate about the patient,
who is translated relative to the x-ray beam (used for localization or patient set-up).
SCREEN-FILM X-ray detector comprising both an intensifying screen and a
film (opposite, NON-SCREEN FILM).

Glossary of terms 35 1
SECOND-GENERATION CT Form of rotate-translate CT using a bank of

detectors (see ROTATE-TRANSLATE CT).
SECONDARY COLLIMATION Usually refers to a simple set of jaws in a
linear accelerator in addition to a multileaf collimator.
SECTION IMAGING Tomography.
SHADED-SURFACE DOSE DISPLAY Computer display of dose. Isodose
surfaces are shown as if they were solid bodies illuminated by light.
SIMULATED ANNEALING An optimization technique which uses princi-
ples modelled on crystal annealing. Such processes converge to a global minimum
result.
SIMULATOR Machine which emulates the geometry of a treatment machine,
but which uses diagnostic energy x-rays to take images of the patient in the
treatment position.
SINGLE-PHOTON EMISSION COMPUTED TOMOGRAPHY (SPECT)
See EMISSION COMPUTED TOMOGRAPHY
SINOGRAM The image formed by stacking projections, one beneath the other,
taken by detectors at sequential (and generally equal) angular increments around
a patient. An off-axis point would describe a sine wave in the sinogram, hence its
name.
SLICE THICKNESS The thickness of a tomographic section.
SMART Stereotactic multiple-arc radiotherapy.
SPILL The burst of charged particles from an accelerator.
SPOT SCANNING Method of creating a large-area proton field by electro-
magnetically changing the position of a small-area beam.
SPREAD-OUT BRAGG PEAK (SOBP) The wide region of high-dose
deposition in the graph of energy deposition versus depth for a charged-particle
radiation. The SOBP is formed by a range-modulating wheel.
STEREOTACTIC RADIOTHERAPY Radiotherapy with the patient’s head
firmly constrained to a precise geometry in relation to the radiation. This may be
by a stereotactic frame bolted to the calvarium, or by some less invasive frame.
STEREOTACTIC RADIOSURGERY As stereotactic radiotherapy, except
the aim is to sterilize the target volume in one or two fractions.
SYNCHRONOUS SHIELDING Method by which a block is suspended in the
field from a rotating treatment machine. The block shields specified volumes of
tissue, whatever the orientation of the gantry.
SYNCHRONOUS FIELD SHAPING Method of conforming the radiation
aperture to the projected size of the target in rotation therapy by gravity-suspended
collimators.
TARGET VOLUME The full extent of the volume to be treated, including the
tumour, marginal spread of disease and a ‘safety’ margin. The ‘biological target
volume’ (BTV) and ‘mobile target volume’ (MTV) are defined in chapter 1. A
new report (ICRUSO) to be published in late 1992 will redefine targets in terms of
‘gross target volume’ (GTV), ‘clinical target volume ’ (CTV) and ‘planning target
volume’ (PTV).

TERMA Total energy released per unit mass. This include all radiation
transport, both particles and photons. Terma has the same dimensions as dose
(compare with KERMA).
THERAPEUTIC RATIO Ratio of the probability of tumour control to the
probability of normal tissue complication.
THERAPY CT SCAN CT scan (for localization purposes) taken with the
patient in the same position as they will be set up for subsequent radiotherapy.
THIRD-GENERATION CT See ROTATE-ONLY CT
THREE-DIMENSIONAL TREATMENT PLANNING SYSTEMS (3DTP)
Computer packages which are able to use 3D image data for planning radiation
therapy, calculating dose distributions and evaluating and displaying these in 3D.
TISSUE INHOMOGENEITY CORRECTIONS Part of the radiotherapy
planning process which takes account of the varying x-ray attenuation of different
body tissues. CT data are often used to assist this process.
TRACKING THERAPY Method of dynamic therapy whereby the radiation
field is constantly adjusted to conform to the projected area of the target. This may
be arranged by moving the patient relative to a stationary source, or vice versa.
The machine at London’s Royal Free Hospital is called the TRACKING COBALT
UNIT.
TRANSIT DOSIMETRY Calculation of the distribution of dose actually
received by a patient, starting with the portal image.
TRANSMISSION COMPUTED TOMOGRAPHY (TCT) X-ray CT as, for
example, distinguished from emission CT (ECT).
TOMOGRAPHY Since 1962the term has been decreed by ICRU to be applied
to all methods of body-section imaging, howsoever performed.
TOMOGRAM Term used for the body section generated by tomography.
TOPOGRAM Same as scout view.
TUMOUR CONTROL PROBABILITY (TCP) The graph of the probability
of killing the tumour as a function of the applied radiation dose.
UNIVERSAL WEDGE Fixed-angle wedge in a linear accelerator. By
combining two fields-one with the wedge and one without the wedg-f
different durations, any arbitrary wedge angle may be achieved.
VIRTUAL SIMULATION Method of creating images of the radiation portals
which will be used at treatment. Instead of these being physically created on
a simulator, they are constructed computationally from tomographic data. The
radiation planning system able to do this is called a VIRTUAL SIMULATOR.
VOLUME OF REGRET A tissue volume receiving more or less dose than
required by prescription.
VOXEL Volume element within the patient, in which some physical property
is measured and displayed.
WEDGE Generally a piece of metal shaped to a wedge with avariable radiation
transmittance along its length. This creates a non-zero angle between each isodose
line and the normal to the beam axis. The same effect can be obtained by moving
accelerator jaws, called a DYNAMIC WEDGE.

WEIGHT Rather loose term, applied to radiation fields delivered by a linear

accelerator, meaning the number of monitor units per field.
WELL-DIFFERENTIATED DISEASE Primary tumour which can be
visualized in images and separated from surrounding normal healthy tissue.
XERORADIOGRAPHY X-ray imaging technique in which a latent image is
formed by the radiation selectively discharging an electrostatically charged plate.
The image is formed by powder deposition.

APPENDIX C
IMPORTANT DEVELOPMENTS IN
‘SUPERVOLTAGE’ RADIOTHERAPY
FOR CONTEXTUAL FRAMING OF
CONFORMAL RADIOTHERAPY
Date Development
1924 The idea of a linear accelerator was first suggested by Gustaf Ising in
Sweden.
1925 Sorenson (California Institute of Technology) created a transformer
capable of producing a peak potential of 1 MV.
1927 Coolidge produced a set of cascaded tubes capable of accelerating
electrons to 900 keV.
1928 Rolf Wideroe first postulated the concept of the betatron accelerator.
1928 Lauritsen used the Sorenson transformer array with a tube to generate
750 keV x-rays at Caltech.
1929 van de Graff built his prototype generator at Princeton.
1930 The first patients were treated with 600 keV radiation from the Lauritzen
x-ray tube at Caltech.
1931 Lawrence accelerated protons to 1 MeV in the cyclotron.
1931 Failla and Quimby at Memorial Hospital (New York) treated patients
with 700 keV radiation from a two-section cascaded Coolidge tube.
1933 Patients were treated with 1 MeV x-rays at Caltech in a laboratory funded
by W K Kellogg ‘the breakfast cereal king’.
1934 van de Graff generator, capable of achieving a potential of 7 MV,
developed at Round Hill, Massachusetts.
1935 1 MeV x-rays were produced from a van de Graaff generator at MIT.
1937 Metropolitan Vickers 1 MV supervoltage unit with Cockroft-Walton
high-voltage supplies installed at St Bartholomew’s Hospital London.
In 1936 the Metropolitan Vickers unit was operated at 700 kV, rising
to 1 MV by 1939. The cost of the unit and building was f12 OOO.
1937 Steenbeck patented the idea for a betatron, but this was not built.
354
Developments in ‘supervoltage’ radiotherapy 355
Date Development
1937 Development of the klystron by the Varian brothers (Russell and Sigurd).
1938 Development of the rhumbatron by W W Hansen at Stanford University.
1939 Randall and Boot developed a magnetron with 1 M W power.
1940 Development of the betatron by D W Kerst at the University of Illinois.
1941 Radioactive cobalt-60 first produced in a cyclotron.
1943 Kerst 20 MeV betatron available for medical use.
1946 Mayneord brought three disks of cobalt-59 to North America for
activation to cobalt-60. TWO were inserted into the AEC Chalk
River reactor and the other was inserted into the Oak Ridge National
Laboratory reactor.
1946 Mitchell (Cambridge) first published a proposal for the use of cobalt-60
as a therapy source.
1948 L G Grimmett (M D Anderson Hospital) produced first practical design
for a cobalt-60 treatment unit.
1948 Although Kent treated one patient in his laboratory with a betatron, the
first clinical radiation programme was started at Saskatoon by Harold
E Johns with an exact copy of the original 2.3 MeV betatron. The
same year, a second Allis-Chalmer’s (A-C) betatron was being used
clinically by John Laughlin at the University of Illinois in Chicago.
1949 William Hansen, linac inventor, died, aged only 40.
1949 First commercially available betatron.
1951 Cobalt-60 radiotherapy started at Saskatoon Clinic (Aug), Ontario
Institute of Radiotherapy (Oct) and at the M D Anderson Hospital,
Houston.
1951 Leksell introduced radiosurgery, initially performed with x-ray equip-
ment.
1952 A home-made cobalt-60 unit was operating at the Los Angeles Tumour
Institute.
1952 The Eldorado Cobalt-60 unit was commercially manufactured.
1953 First medical linear accelerator (8 MV) installed at the Hammersmith
Hospital, London.
1953 Linear accelerators installed at Newcastle General Hospital and at the
Christie Hospital, Manchester.
1954 First (Varian) medical linear accelerator installed at Stanford after early
design work by inventors Hansen and Ginzton.
1954 Vickers 15 MeV linac installed at St Bartholomew’s Hospital Medical
College at Charterhouse Square.
1955 First medical proton irradiation at the Lawrence Berkeley Laboratory,
University of Califomia.
1956 First radiotherapy treatment in the US with a linear accelerator at
Stanford.

356 Developments in ‘supervoltage’ radiotherapy
Date Development
1958 The first treatment-planning systems, which were batch programs with
punch cards and line-printer output, by Wood in Cardiff and Tsien in
Philadelphia.
1959 First conformation therapy with a tracking cobalt unit (London UK).
1959 Gscheidlen’s patent for a prototype multileaf collimator.
1959 Russell Varian died.
1961 Sigurd Varian died in a plane crash off the Pacific Coast of America.
1961 K C Tsien produced a world survey of teletherapy units. There were 1120
worldwide; 47 in Britain (for other totals see Grigg p 3 13).
1962 First Varian clinical accelerator capable of a full 360 degree rotation
installed at the UCLA medical centre.
1962 First electronic megavoltage portal image by Benner.
1965 Takahashi’s famous book on rotation therapy published.
1966 The first interactive treatment-planning system, the Programmed Con-
sole in St. Louis. Development started in 1966-7, but the system was
not available clinically until about 1969-70.
1968 First gamma knife at the Sophiahemmet Hospital, Stockholm, Sweden.
1971 The RAD-8 (Royal Marsden Hospital planning system) was brought into
clinical use in 1971-2 and remained in use for about 15 years.
1980 First modem equipment for electronic portal imaging.
1984 First (modem) multileaf collimator.
This table does not list the many ‘firsts’ in different countries, or by different
manufacturers. Grigg (1965) provides the best comprehensive source of such
information.
REFERENCES
Bentley R E 1992 private communication

Brecher R and Brecher E 1969The rays; a history of radiology in the United States
and Canada (Baltimore, MD: The Williams and Wilkins CO)
Grigg E R N 1965 The trail of the invisible 1ight;from X-strahlen to radio(bio)logy
(Springfield, IL: Charles C Thomas)
Innes G S 1988The one million volt x-ray therapy equipment at St Bartholomew’s
Hospital, 1936-1960 Megavoltage Radiotherapy 1937-1 987 (Brit. J . Radiol.
Supplement 22) 11-16
Jones A 1988The developmentof megavoltage x-ray therapy at St Bartholomew’s
Hospital Megavoltage Radiotherapy 1937-1 987 (Brit. J . Radiol. Supplement
22) 3- 10
Laughlin J S 1989Development of the technology of radiation therapy Monograph
Issue: the technical history of radiology. Radiographics 9 (6) 1245-1266

References 357
Figure C.l. Russell and Sigurd (standing) Varian with an early klystron.
(Courtesy VarianAssociates.)
Sisterson J 1992 (ed) Worldwide charged particle patient totals Particles

Newsletter no 9, January 1992
Takahashi S 1965 Conformation radiotherapy: rotation techniques as applied to
radiography and radiotherapy of cancer Acta Radiol. Suppl242
Varian (undated) Varian Associates: An early history (Palo Alto: Varian Assoc)
FURTHER READING FOR PRE-SUPERVOLTAGE RADIOTHERAPY

Burroughs E H 1986 Pioneers and early years: A history of British radiology
(Alderney: Colophon Ltd)
Mould R F 1980 A history of x-rays and radium (Sutton: IPC Press)
Pallardy G, Pallardy M-J and Wackenheim A 1989 Histoire Illustrke de la
radiology (Paris: Les Editions Roger Dacosta)
Webb S 1990 From the watching of shadows: the origins of radiological
tomography (Bristol: Adam Hilger)

WEBB - The Physics of Three-Dimensional Radiation Therapy

Uploaded by

Copyright:

Available Formats

WEBB - The Physics of Three-Dimensional Radiation Therapy

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

WEBB - The Physics of Three-Dimensional Radiation Therapy

Uploaded by

Copyright:

Available Formats

What is the book about based on the contents?

What is the book about based on the contents?

What developments are mentioned in the table on page 356?

What developments are mentioned in the table on page 356?

Series in Medical Physics

Institute of Physics Publishing

All rights reserved. No part of this publication may be reproduced, stored

British Librav Cataloguing-in-Publication Data

Published by Institute of Physics Publishing, wholly owned by The Institute of

Copyright © 1993 IOP Publishing Ltd.

1 THREE-DIMENSIONAL RADIATION-THERAPY TREATMENT

2 TREATMENT PLAN OPTIMIZATION 65

3 STEREOTACTIC RADIOSURGERY AND RADIOTHERAPY 135

4 THE PHYSICS OF PROTON RADIOTHERAPY 172

5 CONFORMAL RADIOTHERAPY WITH A MULTILEAF

6 MEGAVOLTAGE PORTAL IMAGING 246

7 TREATMENT MACHINE FEATURES FOR CONFORMAL

Copyright © 1993 IOP Publishing Ltd.

8 IMAGING FOR CONFORMAL RADIOTHERAPY PLANNING 306

A NUMERICAL QUESTIONS 337

B GLOSSARY OF TERMS 342

C IMPORTANT DEVELOPMENTS IN 'SUPERVOLTAGE'

Copyright © 1993 IOP Publishing Ltd.

Conformal radiotherapy is tailoring the high-dose volume to the target volume in

Copyright © 1993 IOP Publishing Ltd.

positioning Dose evaluated

Figure P.1. Showing the 'chain' of processes leading to conformal

Copyright © 1993 IOP Publishing Ltd.

Copyright © 1993 IOP Publishing Ltd.

Copyright © 1993 IOP Publishing Ltd.

1.1. CONFORMAL RADIOTHERAPY TREATMENT PLANNING

1.1.1. What is conformal radiotherapy?

Figure 1.1. Illustrates the difference between ‘conventional’ and ‘confor-

vanish. There will still be difficulties associated with incomplete radiobiological

Copyright © 1993 IOP Publishing Ltd.

normalization changed to increase the latter to tolerance and consequently improve

1.1.2. The planning tools required for conformal radiotherapy

1.1.2.1. Conventional 2D treatment planning. Let us recall what we might call

Copyright © 1993 IOP Publishing Ltd.

Figure 1.2. An illustration of some of the features of ‘conventional’ (i.e.

Copyright © 1993 IOP Publishing Ltd.

inhomogeneity corrections were attempted. Many hours at the planning computer

Copyright © 1993 IOP Publishing Ltd.

1 .I .2.3. Three-dimensional multimodality medical imaging and 3 0 treatment

Copyright © 1993 IOP Publishing Ltd.

images for radiotherapy treatment planning (Lichter et all983, McCullough 1983,

Copyright © 1993 IOP Publishing Ltd.

Figure 1.3. An illustration of some of the features of ‘conformal’

Copyright © 1993 IOP Publishing Ltd.

(Goitein et a1 1983b, Kijewski 1987);

0 the University of Tampere, Finland (Vitaenen) (*CADPlan; Varian);

0 the Karolinska Hospital, Stockholm, Sweden (Brahme 1992);

0 the University of Pennsylvania School of Medicine and Fox Chase Cancer

Centre, Philadelphia (Sontag et a1 1987, Sontag and Purdy 1991);

Medicine, St Louis (Purdy et a1 1987, Sontag and Purdy 1991).

I .I .3. The computer system

Copyright © 1993 IOP Publishing Ltd.

1.I .4. Localization and coordinate systems

Copyright © 1993 IOP Publishing Ltd.