X-Ray Micro Tomography
X-Ray Micro Tomography
X-Ray Micro Tomography
John H. Dunsmuir
ExxonMobil Research and Engineering Co.
Annadale, N.J.
[email protected]
INTRODUCTION
tion, and the quality and completeness of the projec-
Reconstructing a two dimensional function from its tion data determine the choice of technique. A Fourier
line integrals taken from many different directions is technique called filtered back projection is the method
the basic process of tomography. When an x-ray beam commonly used in medical imaging. In both medical
is incident upon a specimen several related processes and industrial practice, the ability to image the internal
occur. If the incident x-ray beam completely penetrates structure of a macroscopic object without cutting pro-
the specimen and if the photons scattered or re-emit- vides clear advantages both to diagnosticians and pa-
ted from the interaction can also penetrate the entire tients and to quality control of fabricated parts.
specimen and if refraction effects are negligible, then a X-ray absorption microtomography, referred to as
cross-sectional image of the interaction can be readily XMT or mCT, is the extension of CAT scanning to speci-
reconstructed. Tomographic experiments imaging these mens between 1mm and 1cm in size and creates cross
interactions fall into two broad categories, bright field, sectional images with spatial resolution approaching 1
also called in-line techniques and dark field or off-axis micron. Although the fundamental CT measurement
techniques. In both techniques, the specimen is ro- remains the same for both large and small objects, the
tated to provide the different sampling directions. In a geometry of illumination, x-ray energy range and in-
bright field, the specimen is placed directly between tensity requirements for small objects are well met by
the x-ray source and detector. Spatial resolution is a synchrotron x-ray source. Perhaps most importantly,
obtained by using either a small rastered probe beam the tunability of a synchrotron source can be used to
or a position sensitive detector. Absorption and phase create tomographic images above and below the ab-
contrast are examples of bright field imaging. In dark sorption edge of many elements and provides the abil-
field, a detector is placed adjacent to the specimen ity to make three-dimensional images of the concen-
and inclined with respect to the x-ray illumination axis. tration of those elements.
A probe beam illuminates the specimen and the col-
lected signal represents the line integral of the detected INSTRUMENTATION
interaction along the linear path of the probe beam
through the specimen. The attenuation of the probe Beamline X2-B is designed to measure and recon-
beam is also monitored since dark field data must of- struct x-ray absorption. X2-B is a white beamline with
ten be corrected using an absorption tomogram. Dif- a single crystal monochromator placed in the hutch.
fraction and fluorescence contrast are examples of dark This simple design, built in 1990, was selected to pro-
field techniques. The use of position sensitive detec- vide an approximately 1cm wide by 5mm high illumi-
tors in bright field imaging provides significant speed nated area and to minimize sensitivity to vibration and
advantages over serially sampled dark field methods. source drift. Si<111> and flat multilayer mirrors with
Perhaps the most widely recognized implementa- 2d = 22 and 44Å are available and can provide useful
tion of tomography is medical x-ray Computed Axial x-ray flux between ~7 to ~40 keV. The Si<111> is
Tomography, referred to as a CAT or CT scan. In this recommended for routine use and required for absorp-
imaging technique projections of the two dimensional tion edge crossing. The multilayers provide a nearly
cross-section of x-ray attenuation are acquired from 100 fold increase in incident flux and are capable not
many directions by placing the patient between a con- only of shortening acquisition time but also of damag-
ventional x-ray source and a linear array detector. The ing the detector if not handled properly. Use of the
source-detector pair is rotated about the patient to multilayers must be approved by the PRT prior to ex-
collect the projections from many angles. Reconstruc- periments. Beamline motors for selecting x-ray energy,
tion can be accomplished by a variety of algebraic, sta- crystal position and tilt, and beam height are controlled
tistical of Fourier techniques. The geometry of projec- using the data acquisition computer.
41
ments between 0 and 180 degrees using
Microscope a rotation stage. The number of images
Specimen
Objective needed is about Np/2 where N is the width
of the specimen image in CCD pixels. A
computer is used to control both the
beamline configuration and the
microtomography apparatus. The con-
X-rays
trol software will be discussed in the data
collection section.
Phosphor A second computer performs the re-
Rotation Plate construction process. Each collected 2D
Stage
CCD projection is analogous to a chest x-ray
in medical imaging but is essentially free
Figure 1
of the penumbral broadening and geo-
The geometry of illumination permits the use of a metric distortion associated with source size and dis-
2D position sensitive detector placed directly behind tance to the specimen. As indicated in Figure 1, the
the specimen to simultaneously collect the projection CCD is best thought of as a stack of linear detectors
data for many rays through the specimen. A schematic each associated with a single slice through the speci-
of the detector is shown in Figure 1. The detector men. The distortion free projections make possible
consists of: the reconstruction of perfectly registered 2D slices re-
1. Motorized specimen micropositioners control- sulting in a 3D image. The parallel illumination also
ling specimen height, horizontal translation transverse permits the use of a computationally efficient recon-
to the x-ray beam and specimen rotation. struction technique called direct Fourier inversion that
2. A single crystal scintillator to convert the pat- provides significant speed advantages in reconstruct-
tern of x-ray intensity transmitted by the specimen to ing the 1024 slices of a typical experiment.
a visible light image. A 1cm square by 200micron thick Figure 2 shows the interior of the X2-B hutch.
polished CsI(Tl) crystal is optically isotropic, provides The monochromator crystal in the helium purged, lead
high yield, good stopping power relative to most speci- lined tank to the left of the image. The gantry at-
mens, good spatial resolution, and light output well tached the two theta arm of the monochromator is
matched to the CCD sensitivity. CsI(Tl) provides high levitated on air pads during energy changes. The beam
quantum efficiency, however thin polished crystals are propagates through about one meter of air before it is
not commercially available. The crystal is attached to incident on the specimen and detector located to the
a focussing stage and is moved to refocus the image right side of the image and shown in Figure 3. The
after a lens change. Care must be taken not to collide specimen in this example is a carbon fiber reinforced
the scintillator into the specimen while
refocussing.
3. A microscope objective magnifies the
image onto the surface of a CCD. Image
magnification is set by the objective selec-
tion. 2.5x, 5x, 10x or 20x Zeiss infinity cor-
rected plan apochromats are provided. A
200mm tube lens gives a tube factor of 1.25x.
4. A thermoelectrically cooled Roper
Scientific (Photometrics) CCD collects image
data. The CCD is a back thinned 1024x1024
full frame device with 24micron pixels, ~340K
electron full-well capacity, 14bit digitization,
800kHz readout, 16e/sec dark current at -
35C.
Attenuation images are collected from
many angles by rotating the specimen
through small, evenly spaced, angular incre-
Figure 2
42
PEEK pressure vessel used to study gas evolution dur- where I is the intensity of the absorbed x-ray beam, I0
ing depressurization in an oil saturated sandstone and is the intensity of the incident x-ray beam, r is the
illustrates the relatively large space available for speci- specimen density, x is the specimen thickness, and
men holders. Directly behind the specimen is the scin-
tillator and focussing stage. The microscope objec- m(l) =KZmln
tive, tube lens and turning mirror are internal and not
visible. is the mass attenuation coefficient of the specimen
where Z is the atomic number, m is approximately 4, l
SAMPLE PREPARATION is the x-ray wavelength and n may vary between 2.5
and 3. To absorb 90% of the incident radiation the
Sample selection presumes the presence within the quantity m(l)rx, commonly referred to as t should be
material of heterogeneity that will give rise to detect- approximately 2. To obtain t~2 the user can vary the
able absorption contrast. The specimen must also be specimen thickness and/or the x-ray energy. An in-
dimensionally stable and able to withstand the high x- crease in image noise is readily evident in images re-
ray dose deposited. Specimen heating by the x-ray constructed from projection data where t is less than
beam is usually insignificant. The mechanics of sample 0.5. The detector response becomes non-linear when
preparation are trivial and usually involve simply slic- t is greater than 2.5 and causes reconstruction arti-
ing off an appropriately sized piece of material for place- facts. The strong atomic number and energy depen-
ment on the specimen rotation stage. Specimens are dence provide ample contrast but makes the imaging
usually adhered using tacky wax to a post inserted in a of bulk samples of average atomic number greater than
eucentric goniometer. Epoxy tends to creep and may 20 difficult within the energy range available at X2-B.
compromise specimen stability. Similarly, low average atomic number specimens such
Specimen size selection is driven by straightfor- as polymers and biological materials usually require
ward but sometime conflicting criteria involving absorp- the addition of a dopant to one or more of the compo-
tion, contrast and resolution. The specimen should nents to provide sufficient absorption contrast.
absorb about 90% of the incident radiation along the As mentioned previously, magnification is selected
most radio-opaque path to obtain the best signal to by the choice of microscope objective. Ideally, the entire
noise in the reconstructed image. The absorption of x- specimen should remain within the field of view of the
rays in the 7-40keV energy region covered by X2-B is CCD as it is rotated during a tomography experiment.
described by Clearly, as the magnification is increased the field of
view decreases and the specimen dimensions must be
I/Io = e-m(l)rx adjusted accordingly. For some specimens, the size
required by the resolution criteria does not permit t ~2
within the energy range available at X2-B. In
other cases t ~2 can only be obtained at x-ray
energies above 20keV resulting in some resolu-
tion loss due to the finite stopping power of the
scintillator.
If the specimen width remains completely
within the field of view during CT then the recon-
struction will be a quantitative map of linear at-
tenuation coefficient. If the specimen or parts of
the specimen container extend beyond the field
of view, reconstruction is still possible using local
techniques, however the resulting images will have
a relative grayscale in arbitrary units.
The selection of specimen size, x-ray energy
and resolution can also be driven by the differ-
ences in x-ray absorption among the different
components of the specimen. For example, a
specimen where resolution requirements are
modest and mass absorption is low may indicate
Figure 3
43
using a relatively large specimen size and high inci- cedures. Similar dialogs are available for controlling
dent x-ray energy. If the contrast among the compo- the beamline motors, changing the x-ray energy, run-
nents of the sample is low then a lower x-ray energy ning a tomography experiment, sampling the tomog-
and smaller sample size will probably be needed. raphy data and interactive reconstruction of single or
If a difference tomogram is to be acquired across small groups of slices.
an absorption edge then the x-ray energy and there- For example, as discussed previously, initial radi-
fore specimen size is fixed. The lower limit of determi- ography is needed to determine the specimen x-ray
nation in a 3D image is about 1000ppm for many ele- attenuation if the sample is of unknown composition
ments provided that t~2 at the absorption edge en- and structure. This is done by first determining the
ergy. If a difference tomogram is to be acquired for exposure time for the incident x-ray beam at the initial
more than one element then the size selection is a estimated energy and then collecting a TauMap of the
compromise depending on the concentration, edge specimen using the menu option. TauMap collects ra-
positions and jump ratios of the elements of interest. diographic images with and without the sample in the
The reconstruction software requires that projections beam and calculates the log of I/Io. The specimen is
to be subtracted for difference tomography must be in moved during the TauMap process and a field is pro-
the same format. vided in the dialog box for the user to enter how far to
Selection of the appropriate specimen size usually move the specimen. Using this dialog the user can
involves some a-priori knowledge of the specimen com- determine an exposure time, magnification, x-ray en-
position and structure. Spreadsheets for calculating ergy and sample size appropriate for the specimen.
the mass attenuation coefficient of compounds and the Prior to a CT scan the specimen should be manu-
attenuation of samples of given size are available at ally aligned on the eucentric goniometer to minimize
the beamline. For true unknowns, a series of radiogra- specimen wobble about the rotation axis. Also the ro-
phy experiments is needed to determine the appropri- tation stage axis must be aligned so that it is exactly
ate specimen size based on its x-ray attenuation. Al- parallel to the CCD columns and perpendicular to the
though there is considerable latitude in selecting speci- incident x-ray beam. Automations such as Find Axis,
men size, finding the optimum imaging conditions may ScanAxis and Align Check assist the user in evaluating
require several adjustments. the degree of misalignment. The user enters correc-
Lastly, the design of specimen holders such as en- tions using the beamline and CT stage motors. Align-
vironmental or mechanical cells is the responsibility of ments are discussed further during beamline training.
the user who must take into consideration the x-
ray absorption properties and geometry of the cell
materials. Although a relatively large area is avail-
able for specimens, best absorption imaging is
obtained when the specimen is in close proximity
to the CsI scintillator.
DATA COLLECTION
The user controls the function of the
microtomography system through the data acqui-
sition computer running a PRT extended version
of IPLab, a commercial 2D image processing pro-
gram. The PRT extensions to IPLab that provide
Figure 4
motion control and image acquisition are available
from the Ext. menu. The details of the acquisition
software are discussed in documentation available at X2-B supports several CT scanning modes:
the beamline and are presented during user training. 1. CCD binning and ROI acquisition, useful for
The examples shown below are for information only quick low-resolution scans.
and may differ slightly from those currently in use at 2. Global scanning, provides quantitative images
X-2B. An example dialog box for simple image acqui- of linear attenuation coefficient when the entire width
sition is shown in Figure 4. The popup Expose menu of the specimen remains within the field of view during
provides several kinds of exposure and alignment pro- rotation.
44
3. Local scanning, provides qualitative grayscale for images of specimens that extend beyond the field
images when all or part of the specimen extends be- of view of the CCD can be considerably longer than
yond the field of view during rotation. Specimens should that used for the unattenuated beam since all points in
be no more than 3x the detector field of view. the image are attenuated by the sample. Similarly, a
4. Field doubling, the axis of rotation is placed by beam shaper is available to match the pattern of inci-
the user at either the right or left edge of the field of dent intensity to the roughly cylindrical specimen shape
view and the specimen is rotated between 0 and 360 allowing nearly constant count statistics across the field
degrees. Field doubling collects 2048 wide data from of view. It is important to collect as many counts in a
the 1024 wide CCD. single calibration and image frame as the CCD dynamic
5. Beam shaping for both global and local scans, range will allow. These counting statistics have direct
the incident intensity profile is adjusted to provide nearly impact on the signal to noise of the reconstructed slices.
constant counting across the width of the specimen Scan times can vary considerably depending on
image. the flux available from the synchrotron at a particular
6. Correlated sampling, improves S/N and re- x-ray energy, the image magnification, the number of
moves most systematic errors from low contrast speci- view angles needed, and the signal to noise required in
mens. The specimen exposure time can be decreased the reconstructed volume. A low magnification 256
by a factor of 2 with only a slight increase in noise cube of data at about 20keV can be acquired in a few
compared to normal sampling. minutes. A single 1024 cube of data may need from 1
The basic CT scanning protocol is automated. A to 7 hours. The longest protocol supported will ac-
scan begins by translating the specimen out of the x- quire a 2048x2048x1024 data volume using correlated
ray beam and summing several Io beam images into a sampling to reduce noise and usually requires the use
calibration buffer. The specimen is placed back into of a multilayer mirror to provide sufficient flux to com-
the beam and images are acquired at small angular plete during a single x-ray ring fill.
increments between 0 and 180 degrees. Since the An externally mounted SCSI or FireWire drive pro-
synchrotron ring output decays with time, the user may vided by the user and attached to the reconstruction
elect to repeat the calibration procedure periodically computer is recommended since the data sets are suf-
during the scan. The acquisition program computes ficiently large to preclude practical transmission over
6000* ln(I/Io) for each specimen image and concat- current networks. Data transmission between the ac-
enates the result as a 16bit integer to a single file on quisition and reconstruction computers is asynchro-
nous and causes no loss in acquisition speed for expo-
sure times greater than 2 seconds.
IPLab provides a scripting utility that allows the
user to construct extended or complex CT scanning
procedures. For example, scripts that execute a series
of CT scans and specimen height changes can be used
either to scan multiple specimens or to create 3D vol-
umes extended in the specimen height direction. Al-
ternately, beam energy can be changed for mapping
certain elements. The beamline is capable of extended
unattended operation using IPLab scripts, however the
instrument does not currently monitor the incident
beam intensity.
DATA ANALYSIS
45
corresponding 2D slice. A read function is available in of the specimen. Analysis of these images to obtain
IPLab to read projection data, reformat it into a image metrics relevant to the users research is, strictly
sinogram and display it on the screen. The sinogram speaking, beyond the purpose of the beamline. X2-B
can then be reconstructed by selecting the parameters does provide a limited set of 3D analysis and visualiza-
in the dialog box shown in Figure 6. As shown, the tion tools that can be made available to the user for
sinogram will be reconstructed with a rotation axis at use at the beamline. Further analysis can be done at
column 128 without any adjustments or filtering using the Center for Data Intensive Computing at BNL or at
a global direct Fourier algorithm. The dialog contains the users home facility using commercially available
options for conditioning the data to correct common analysis, visualization and development packages such
artifacts and for region of interest reconstruction. as Iris-Explorer, MatLab, AVS Express, IDL, PVWave,
A free-standing application, CT_Recon, uses the and 3dma(SUNY). The PRT does not endorse or sup-
parameters determined interactively in IPLab to recon- port these packages but can supply file format specifi-
struct the entire 3D volume. The main dialog for cations to aid in importing the images acquired at X2B.
CT_Recon requests the same information as shown in
figure 6 and has additional fields for selecting the out- EXAMPLES
put format and range of slices. CT_recon supports
drag-and-drop, import of tomography data in other In this example we acquire a 1024 cube of data of
formats, batch processing of multiple projection files a sand pack and extract statistical properties of the
and several output formats. pore space. The sand pack was prepared by pouring
The final product of the reconstruction process is a Ottawa sand, a common laboratory reagent, into a 1cm
single file containing a sequence of contiguous CT slices ID polyethylene tube. A 10mm/pixel pitch using the
2.5x lens was sufficient to image nearly the entire ID
of the tube while capturing the pore shape and con-
nectivity of this coarse-grained sand. The vertical di-
mension of the acquisition is limited by the beam height
available at X2-B to about 5mm. If required the speci-
men height can be moved and 3D volumes concat-
enated to a vertical dimension of nearly 10cm. An x-
ray beam energy of 25keV gave t ~2 in the projection.
The axis of rotation is at pixel column 512. The local
protocol was used to acquire and reconstruct the pro-
jection data since the tube and a small portion of the
Figure 6 Figure 7
46
sand pack extended beyond the CCD field of view. A tion shown in Figure 8 is the probability of finding a
256 cube was extracted from an arbitrary position within chord of length L in the 3D volume. The two-point
the 3D data and was visualized using VoxBlast, a com- probability function shown in Figure 9 is the probabil-
mercial visualization package loaded on the reconstruc- ity that two points selected randomly in the volume
tion computer. The 256 cube was segmented into pore will both lie in this case in the pore phase. The asymp-
and solid using a simple grayscale threshold and ren- tote of S(r) describes the correlation length for the
dered so that the solid sand grains are transparent pores, in this case about 250mm. These measures are
revealing the 3D shape of the pore space as shown in related to transport, diffusion and trapping in random
Figure 7. 3DToolKit is a PRT extension to IPLab for porous media.
analysis of 3D images. It contains tools for segmenta-
tion, connectivity, and statistical analysis of 3D images.
In this example, we extract two statistical measures,
USEFUL WEBSITES
the chord length and two point probability distribu-
tions from the full 3D data. The chord length distribu- IPLAB
http://www.scanalytics.com/
1
O t t o w a S and Cho rd Le ng t h D is t rib ut io n
CCD CAMERA
http://www.roperscientific.com/
0 .1
IRIS EXPLORER
0 .0 1
http://www.nag.co.uk/Welcome_IEC.html
P(L)
0 .0 0 1
AVS EXPRESS
0 .0 0 0 1
http://www.avs.com/
-5
10
MATLAB
10
-6
http://www.mathworks.com/
0 200 400 600 800 1000 1200 1400
L(mic ro ns )
PVWAVE
Figure 8 http://www.vni.com/products/wave/index.html
VOXBLAST
http://www.vaytek.com/
Two P o int P ro babilit y Dis t ribut io n
0.3
CDIC
0 .2 5 http://www.bnl.gov/cdic/Sci_Projects/Computer_Sci/
Visualization/visualization.htm
S(r)
0.2
3DMA
0 .1 5 http://www.ams.sunysb.edu/~lindquis/3dma/
3dma.html
0.1
IDL
0 .0 5
0 10 0 200 3 00 40 0
R(µm)
500 6 00 70 0 http://www.rsinc.com/idl/index.asp
CT Book
Figure 9 http://www.slaney.org/pct/
47
USEFUL REFERENCES
Developments in X-ray Tomography, Ulrich Bonse Ed.,
Proceedings of the SPIE, vol 3372 and 3149
REPRESENTATIVE PUBLICATIONS
48