Reconstructing 3-D light-microscopic images by digital

image processing

Angelika Erhardt, G. Zinser, D. Komitowski, and J. Bille

A reconstruction procedure based on linear system theory has been developed for 3-D light-microscopic im-
ages. Inverse filtering with the 3-D optical transfer function was used for imagereconstruction. The proce-
dure allows a significant improvement in spatial resolution in the image planes perpendicular to the optical
axis.

1. Introduction was derived and is used for the inverse filtering proce-
In many applications of light microscopy, high reso- dure of the image.
lution objectives with a low focal depth have to be used, Provided that the integral absorption within the
while, usually, the size parallel to the optical axis of the object is low, the method is object independent and is
object under investigation exceeds the depth of focus. therefore not restricted to any special application. In
Consequently, a light-microscopic image is actually an our laboratory this procedure was applied to various
optical section of the object at the focal plane. It con- histological objects. The results obtained show a sig-
tains only a fraction of the total information content nificant improvement in the image quality, so that fine
present in the object. In addition, superimposed on structures of the object become visible, and the image
such an image are defocused projections from adjacent can be used for more sophisticated measurements.
focal planes. The result is a low spatial resolution
within the image. By 2-D image processing these dis- 11. Theory
tortions cannot be eliminated. The light-microscopic image of an absorption-free
The only appropriate way to attack this problem is 3-D object f(x,y,z) can be described in terms of linear
to generate 3-D images. In contrast to the 2-D case, a system theory: the 3-D imageg(x,y,z) is generated by
3-D light-microscopic image contains the maximum a convolution of the object f(xy,z) with the 3-D point
available information. It can be recorded as a series of spread function psf(xy,z) of the image forming
2-D images by varying the focus plane of the objective. system:
Although each of these image planes is still superim-
g(x,y,z) PSf(x,y,z).
posed with projections from the whole object, they can (1)
be eliminated and the original object can be restored. Let G(u,v,w) and F(u,v,w) be the Fourier transforms
The purpose of this paper is to present a new recon- of g and f, respectively,and OTF(u,v,w) the 3-D optical
struction algorithm for 3-D light-microscopic images. transfer function of the microscope. Then, in the
It is based on the mathematical formulation of the 3-D Fourier domain, the convolution reduces to a simple
imaging properties of the light microscope in the multiplication:
framework of linear system theory. The 3-D optical
G(u,v,w) = F(u,v,w) *OTF(u,v,w).
transfer function of the transmission light microscope (2)
Therefore, in the absence of noise, the operation
F(u,v,w) = G(u,v,w)/OTF(u,v,w) (3)
carried out in the region of definition of the optical'
J. Bille is with University of Heidelberg, Institute of Applied
transfer function and the inverse Fourier transforma-
Physics I, D-6900 Heidelberg, Federal Republic of Germany; the other tion of F generate the reconstructed object.
authors are with German Cancer Research Center, Institute of Ex- Because of the nonlinearity of light absorption, linear
perimental Pathology, D-6900 Heidelberg, Federal Republic of system theory requires absorption-free objects. In
Germany. practice, of course, microscopic objects are never free
Received 29 June 1984. of absorption, but it is sufficient that the absorption
0003-6935/85/020194-07$02.00/0. within the object is small enough to be approximately
© 1985 Optical Society of America. linear. In this case the linear system theory can be
194 APPLIED OPTICS / Vol. 24, No. 2 / 15 January 1985
 in the z direction, the direction of the optical axis. This
generates-the 3-D optical transfer function OTF(q,s):

q =Os = 0,
2
2 f(q) [ S 11/2
OTF(q~s)={ 1 - h- _
h[IL)] 0< q < 1, Isl h(q),
7r0zoqoh(q)
elsewhere,

(6)

Fig. 1. Two-dimensional radial section of the 3-D optical transfer

function. The color code is based on a logarithmic scale: white where z0 =-extension of the 3-D image in the z direction
corresponds to the value 1, black to values <10-3. The spatial (normalization constant), and
frequencies are normalized to the incoherent cutoff frequency qo. s = ii/qO.
The calculation of OTF(q,s) for this figure is based on magnification Figure 1. shows a 2-D section of OTF(q,s) at v = 0.
of the objective, 10OX;N.A., 1.3; wavelength of light, 0.53 m. Rotation about the w axis yields the complete 3-D op-
tical transfer function.
applied and an optical transfer function can be calcu- Because of the symmetry of the 2-D transfer func-
lated. Most histological or cytological objects fulfill this tions, OTF(q,s) is radially symmetric with respect to the
condition. w axis, which corresponds to the optical axis. It forms
The 3-D optical transfer function OTF(u,v,w) can be a strong low pass, and, especially on the w axis it is a 3
evaluated from the 2-D optical transfer functions function. It is bandlimited in all directions and van-
OTF(u,v;z) for various degrees of defocus z, which can ishes outside a tQruslike region. The cutoff frequency
be derived from scalar diffraction theory.' An ap- within the (uv) plane is identical with the incoherent
proximation of the 2-D optical transfer functions given 2-D cutoff frequency q0;the maximum cutoff frequency
by Stokseth2 was used, together with another approxi- in the w direction is N.A./4 and is located at position
q = 1/2. Figure 2(a) shows the values of OTF(q,s) on the
mation, giving the dependence of defocus z in the object
space on the maximal phase error u' at the exit aperture q axis and Fig. 2(b) parallel to the w axis at q = /2.
of the objective Figure 2(c) shows the values of the 3-D optical
2
transfer function for w = 0 in comparison with the in-
a= (N.A.) z/2. (4) focus 2-D transfer function, to illustrate the low pass
(For detailed derivation of this approximation see the character described. The 2-D in-focus optical transfer
Appendix.) With these two approximations the 2-D function can be obtained from the 3-D optical transfer
optical transfer functions can be expressed in the function by integration parallel to the w axis.
form The uniqueforin of the 3-D optical transfer function
is determined by the physical properties of the micro-
OTFfqz) = 2f(q) -J 1[2-rh(q)qoz] (5) scope: the band limitation parallel to the w axis is a
27rh(q)qoz result of the infinitely extended light cone in the di-
where f(q) = 1 - 1.38q + 0.03q
2
+ 0.344q 3 , rection of the optical axis. The width of the transfer
h(q) = N.A. q(l-q), function depends on the numerical aperture of the ob-
2 2 12
q = (U + V )' /qo, jective, which is-correlated to the angle of aperture of
Jl(x) = first-order Bessel function, the light cone. The 6 function on the w axis, corre-
qo = incoherent cutoff frequency, and sponding to the optical axis, expresses the constancy of
N.A. = numerical aperture of the objective. the light intensityi.e., is a result of energy conservation.
A detailed examination of the approximations indi- The extreme low pass formed by OTF(q,s) in compar-
cated that the above expression is only valid within a ison with the 2-D in-focus transfer function is the cause
limited defocus distance, i.e., the 3-D object can only of contrast reduction in the 3-D image. It is induced
have a limited extension parallel to the optical axis. by low-frequency projections of defocused regions of the
This length depends on the numerical aperture of the object situated in adjacent planes to the focus level.
objective. Representative values are: These projections cause a relative suppression of me-
dium and high frequencies.
16 Aumfor an objective 100X/N.A. 1.3 Because of the band limitation of the 3-D optical
200 Aimfor an objective 25X/N.A. 0.65 transfer function, it is impossible to completely recon-
1000 ,im for an objective 10X/N.A. 0.25. struct the object ffom its 3-D image. The image for-
In most cases the size of the objects investigated is mation process of a ight microscope is accompanied by
within these limits. an inevitable loss of information on the object. It is
The 2-D optical transfer functions are symmetric with impossible to regain this information, regardless of the
respect to the origin. Within the above limits they are reconstruction procedure used. Within the margin of
independent of the direction of focal shift. OTF(q;z) error due to the noIinearity of the light absorption, the
has an analytical form and can be Fourier transformed result obtained by inversefiltering of the imagewith the

15 January 1985 / 1. 24, No. 2 / APPLIED OPTICS 195

 0 -

Table 1. Cutoff Frequencies q0 and wo1qoand Maximum Sample

Spacings6-Y and 6, Perpendicularand Parallel to the Optical Axis for
a Various Objectives
G -1
Objective 100X/1.3 25X/0.65 1OX/0.25
I7 qo(1/gm) 4.91 2.45 0.94
I
-2 wo/qo 0.33 0.16 0.06
6." (AM) 0.102 0.204 0.53
6a(m) 0.31 1.28 8.48
OR
- I , Il
0.0 0.2 0.4 0.6 0.8 1. 0
q

-2 struments, Inc.) The maximum sample spacings per-

pendicular and parallel to the optical axis are deter-
mined by the sampling theorem of the discrete Fourier
V; transformation and the cutoff frequencies of the 3-D
S -3
I,- ' optical transfer function. Table I shows the cutoff
I
13 frequencies q0 and wo/q0 perpendicular and parallel to
the optical axis and the corresponding maximum sam-
pling spacings 6by and &,in both directions for three
different microscope objectives.
0.0 0.2 0.4 0.6 0.8 1. Data acquisition is controlled by a PDP11/34 pro-
cessor and the data are stored on magnetic tape for
subsequent off-line treatment. To avoid errors re-
sulting from instabilities of the microscopeillumination,
a series of background images is collected in parallel to
the focus series for calibration purposes. Oversampling
G-
and subsequent reduction of the images allows reduc-
tion of the influenceof the transfer function of the video
camera, which acts as a low pass filter on frequencies
perpendicular to the optical axis. The images are ov-
ersampled by a factor of 3 and subsequently reduced by
0.0 0.2 0.4 0.6 0.8 1. I the same factor by calculation of the mean grey values.
This procedure has the additional effect of increasing
- the signal-to-noise ratio by a factor of 3. For further
j, improvement every image plane of a series is sampled
1 several times (three to five) and the average of the im-
10
ages is calculated.
-I

IV. Reconstruction
V C 0
_
The reconstruction of the focus series is performed
0- off-line on a computer VAX 11/780 with a 2-Mbyte
0.0 0.2 0.4 0.6 0.8 1.0 memory.
q
The values of the 3-D optical transfer function are
Fig. 2. (a) One-dimensional section of the 3-D optical transfer
function on the q axis (s = 0); (b) parallel to the s axis at q = 1/2. (c)
very low near the marginal surface of its bandpass vol-
Values of the 3-D optical transfer function at w = 0 (solid line) com-
ume. In the spectrum of the image, the nonzero values
pared with the 2-D optical transfer function for zero defocus (dashed of the corresponding frequencies are mainly due to
line). (d) The inverse optical transfer function (dashed line) and the noise. Dividing these frequencies by the low values of
inverse effective optical transfer function vs the normalized radial the optical transfer function would cause an overam-
frequency q. All the calculations are based on objective magnification plification of noise, which is certainly not desired.
100X, N.A. 1.3, light wavelength 0.53 Atm. Experiences with various 3-D test objects have shown
that an effective inverse optical transfer function of the
following form yields very good results:
3-D optical transfer function contains the maximum
information about the object which is left after the OTF-l(q,s) if OTF'1(q,s) 100,
image formation process.
OTF-'(q,s) = OTF'l(q,s) cos[p(q,s)] if 100 OTF 1l(q,s) 1000,
III. Data Acquisition 0 elsewhere,
To generate 3-D light-microscopic images we use an
Axiomat microscope (Zeiss) and a plumbicon scanner.
The video signal is digitized into 64 grey levels by an (7)
image analysis system, Quantimet 720 (Cambridge In- wherep(q,s) = log[OTF-'(q,s) - 2r/2.
196 APPLIED OPTICS / Vol. 24, No. 2 / 15 January1985
 The maximum of OTF-l(q,O) occurs at -60% of the
incoherent cutoff frequency qo. In Fig. 2(d) the values
of OTF-A(q ,O) are shown in comparison with the values
of the inverse optical transfer function.
The effective inverse optical transfer function still
forms a high pass filter, which causes high positive and
negative values at the edges within the image. In cases
of high edges at the boundaries of the image they can be
suppressed by multiplication of the image with a 3-D
window function.3 For this purpose a Hamming or
Hanning window can be applied. Windowing the image
with a Hamming or a Hanning function before the re-
construction procedure is equivalent to a convolution
of the image spectrum with the Fourier-transformed
window function. In the case of the Hamming or
Hanning window the Fourier transform is very similar
to a 3 function. So the error caused by windowing is
relatively small compared with the error introduced by
overamplification of the boundary edges.
The 3-D preprocessed image is transformed into the
frequency domain with the 3-D Cooley-Tukey algo-
rithm. 4 In the limit of negligible absorption, the fre-
quency spectrum of the image should vanish outside the
region of definition of the optical transfer function. In
fact, however, this part of the spectrum contains low
frequency-independent randomly distributed values
caused by noise. These values are replaced by zero
before inverse filtering with the optical transfer func-
tion.
After multiplication of the frequency spectrum of the
image with OTF-1(q,s) and inverse Fourier transfor-
mation, the reconstruction procedure is completed. In
a postprocessing step the inverse Hamming or Hanning
window is applied, depending on the preprocessing step
chosen. The multiplication with the inverse window
is only carried out within the inner part of the image.
V. Results
Three-dimensional images from Feulgen stained cell
nuclei and various test objects with known 3-D struc-
ture, such as radiolarians and bristles from insect wings,
have been recorded as focus series and reconstructed
according to the procedure described above. Figures Fig. 3. Nucleus of a cell from the mucosa of the large intestine:
3-5 show three examples of the results of the recon- diameter of object, 8 Mm; objective, 100X; N.A. = 1.3, oil; total focal
struction procedure. The reconstructed planes of each shift parallel to the optical axis, 16 lm; number of averaged images,
3-D image (right) are compared with the corresponding 3; factor of oversampling and subsequent reduction, 3; sample spacing
planes of the original series (left). perpendicular to the optical axis, 0.125 pm (after reduction); sample
Figure 3 shows the nucleus of a cell from the mucosa spacing parallel to the optical axis, 0.25 pm; preprocessing steps, none;
of the large intestine. The main histological features postprocessing steps, none. (a) and (b) Planes perpendicular to the
of the nucleus (i.e., nuclear membrane, nucleolus, and optical axis. The original images show an optical dense region at
condensed chromatin granula) are well preserved in the coordinates x = 38, y = 28 which has a clearly visible fine structure,
reconstructed image. The nucleolus can be seen in the as can be seen in the reconstructed image. (c) and (d) Planes parallel
to the optical axis. The optical dense region is located at (c) coordi-
image planes perpendicular to the optical axis [Figs. 3(a)
nates y = 28, z = 18 and at (d) coordinates x = 38, x = 18. The de-
and (b)] of the original focus series as a region of rela- focused projections in the original images are apparent, as well as the
tively high optical density, located close to coordinates enhancement of the spatial resolution perpendicular to the optical
x = 38, y = 28. This region looks similar in both planes. axis. (d) A 1-D section for y = 28, z = 18. The optical density at x
In the reconstructed series, however, the fine structure = 32 - x = 40 in the original image shows no indication of a gap, while
of this region is clearly visible. It is subdivided into two in the reconstructed image two maxima are visible. This shows that
smaller parts lying in close proximity, oriented almost the reconstruction procedure does not merely amplify already existing
parallel to the optical axis. Comparison of the two maxima and minima in grey values but removes defocused projections
planes z = 15 and z = 17 shows that the two regions in the 3-D image.

15 January 1985 / Vol. 24, No. 2 / APPLIED OPTICS 197

 2 .2 0 4t

o7~ .2 -
-e

.0 0 D.

9:.15
HiUEA |;

C0 ,, 0 i

C C90_w

60 ' T : Fig. 5. Test object, crossing bristles of an insect wing: diameter of

I. X X 0^-o object, 1 m; objective, 10OX;N.A. = 1.3; oil; total focal shift parallel
z .: . a .^ to the optical axis, 16 m; number of averaged images, 3; factor of
* r.0 "a
of oversampling and subsequent reduction, 3; sample spacing perpen-
C' 0
,°. cot .; no dicular to the optical axis, 0.125 m (after reduction); sample spacing
>^ A 3 ok >0 parallel to the optical axis, 0.25 Am; preprocessing steps, Hanning
bM Cd window; postprocessing steps, inverse Hanning window. (a) and (b)

*°^0; =,
s X a;°Planes perpendicular to the optical axis; (c) plane parallel to the op-
X * tical axis. The reconstruction of a test object with known 3-D ge-
E ,.t s F ometry ensures that the structures in the reconstructed images are
> _ 0 not artifacts but fine structures of the object.

~ 0- O

0
2- . o reveal a screwlike structure parallel to the optical axis.
0 I, Fig. 3(c) shows a section parallel to the optical axis at
2 the coordinate x = 38. The improvement of the spatial
V0a:o resolution perpendicular to the optical axis is apparent.
go) The fine structure is clearly visible in the reconstructed
X r as d image, whereas in the original series it is distorted by
:d i, *3 defocused projections from adjacent object planes.
o., . Parallel to the optical axis, only slight improvement of
the spatial resolution can be achieved. Figure 3(d)
CZ 0 shows an image plane parallel to the optical axis at y = 28
-0 > together with a 1-D section at z = 18. As can be seen
AE;0 . from the distribution of grey values in this section, the
.Z ) high absorbing region in the original focus series shows
0 dQ = no indication of a separation at this position. This
confirms that the separation of the two regions in the
. Q; < reconstructed image is not merely an amplification of
bin
> 0if an already existing gap in the distribution of grey values
. 0 M but in fact a separation of two regions.

198 APPLIED OPTICS / Vol. 24, No. 2 / 15 January 1985

 Figure 4 shows a reconstruction of a focus series of a form of the 3-D optical transfer function derived on the
radiolarian (Radiolaria Aulonia hexagonia). Figures basis of the linear system theory. Studies with such a
4(a) and (b) show two planes perpendicular to the op- modification of the optical transfer function are being
tical axis from the front hemisphere. In Figs. 4(c) and carried out at present in our laboratory. The function
(d) two planes from the back hemisphere of the object is object dependent and utilizes the information con-
can be seen. The reconstructed series give a clear re- tained in the part of the Fourier spectrum of the 3-D
production of the details of the hexagonal lattice of the image lying outside the region of definition of the 3-D
radiolarian shell, which in the original series is super- optical transfer function.
imposed with defocused projections. Parallel to the Agard and Sedat5 and Mathog et al. 6 recently re-
optical axis only the outlines of the hexagonal lattice are ported a different reconstruction procedure which was
visible. It is elongated along the optical axis. The applied to optical fluorescence microscopy to study the
reasons for this are the limited bandwidth of the 3-D chromosome topography of Drosophila polytene nuclei.
optical transfer function parallel to the optical axis and They used an iterative method in which, from an initial
the resulting information loss in this direction as de- guess at the object (i.e., the image), a self-consistent
scribed above. solution is generated. Each image plane is deblurred
Figure 5 shows the reconstruction of a test object with by successive inverse filtering with the 2-D optical
known 3-D structure: two crossing bristles of an insect transfer function for 2n adjacent planes, derived by
wing. In the image planes perpendicular to the optical Stokseth.
axis at the crossover point of the two bristles, the orig- Our approach described here is a straightforward
inal series shows a significant enhancement of the op- deconvolution based on the 3-D physical imaging
tical density: the projection of the out-of-focus bristle, properties of the light microscope. Consequently it
which is missing in the reconstructed series. However, permits a more exact reconstruction. No enhancement
a small part of the projection is still visible in the re- techniques are necessary in a postprocessing step,
constructed planes. Since the objects are low-absorbing making the result more reliable. The only requirement
and have a cylindrical form, the boundary lines have a is that the integral absorption of the object lies within
higher absorption than the middle area of the object. the tolerable limits of linear system theory, i.e.,that the
Therefore, in the planes parallel to the optical axis [Fig. absorption can be considered linear. This restriction
5(d)] only the two boundary lines of the object are however, must also be fulfilled by the iterative method.
visible. The objects in the reconstructed series are The CPU time required for the exact reconstruction
clearly separated, whereas in the original series they are proposed here is similar to that of the iterative
connected by defocused projections. method.
The method for the reconstruction of 3-D light-mi-
VI. Discussion croscopic images reported here was developed within
The method of 3-D reconstruction of light-micro- the framework of studies about quantitative description
scopic images described above is object independent possibilities of qualitative histological features of cell
and results in an improvement in spatial resolution, nuclei but can be applied to all other light-microscopic
especially perpendicular to the optical axis. Fine imaging problems if the restrictions concerning object
structures which are normally superimposed with de- size and absorption level are fulfilled.
focused projections become visible, allowing more ac-
curate measurements. The only prerequisite for the Appendix: Derivation of the Approximation a'
2
reconstruction method is a low overall absorption within (N.A. ) z12
the object, since the method is based on linear system Hopkins's formula' as well as Stokseth's approxi-
theory. On the other hand, low absorption implies a low mation 2 of the 2-D optical transfer functions for various
contrast and a low signal-to-noise ratio. Between these focal shifts are given in dependence of the maximum
two extremes, high signal-to-noise ratio and minimum phase error a' at the exit pupil on the image side of the
absorption, a compromise has to be found. This limits optical axis [Fig. 6(a)]. Practically though, this value
the set of objects to which the method can be applied. cannot be measured. Instead the focal shift z [Fig. 6(b)]
Previous work on histological preparations has shown on the object side can be taken from the focal position
that in many cases such a compromise is attainable.. of the microscope. Therefore it is necessary to derive
Even though the spatial resolution perpendicular to the relation between a' and z.
the optical axis can be considerably improved, the From Fig. 6(a) the following equation is obvious:
spatial resolution parallel to the optical axis is not sig-
(r' + Z')2 = (r' + o')2 + Z'2 - 2(r' + a')z' -cos(wr
- a'), (Al)
nificantly higher in the reconstructed image than in the
original image because the optical transfer function which can be rewritten
parallel to the direction of the optical axis is limited to
a small bandwidth. a' = -r' - z' cosa' + (r' + z')(1 - e)1/2, (A2)
If the absorption of the object is not extremely low, where
an enhancement of the spatial resolution parallel to the 2 2
Z' sin al'
optical axis seems to be possible. For this purpose, an E (r + Z')
2

experimental transfer function has to be found which,

in the limit of vanishing absorption, has to match the With the well-known relations

15 January 1985 / Vol. 24, No. 2 / APPLIED OPTICS 199

 N.A. 2 z = sina' sinf'z', (A7)
where

I a = arctan (r' +

A = r' tana' [Fig. 6(b)].

From Eq. (A7) the expression for z' cannot be formu-
lated in algebraic terms. Therefore, the values for z'
N-
in dependence of z were calculated numerically. The
values of the inverse function z for various z' were
object achieved by an exchange of z and z'. They can be ap-
proximated numerically by a fifth-order polynomial
where the first coefficient ao vanishes, with the least-
squares error fit method:
5
b Z, = E aiMi+izi (A8)
i=1
T 2
= zM [1 + f(z)] (A9)
and Eq. (A6) can be rewritten
a' = N.A.2z[1 + f(Z)][1 - g(Z)], (A10)
2
Z Z Z opt axis where
r ra
5
object side I imageside f(z) = (al - 1) + E aiMi-lzi-l
i=2
Fig. 6. (a) Defocused optical system on the image side: A denotes 5
the radius of the exit pupil; r' the focallength; z' the focalshift on the E aiMi+lz
optical axis; a' the aperture angle. The maximum phase error is given g(z) i=l
by the difference of path lengths of the rays at the edge of the aperture 5
as a result of a focal shift. (b) Object and image side of an optical r' + E_ aiMi+lzi
i=1
system for positive and negative focal shifts.
Yortile limits of the local shits given in this paper t(z)
and g(z) are approximately zero, and the relation be-
n sina = M sina', (A3) tween the maximum phase error d' and the object-sided
(A4) focal shift z can be expressed by
N.A. = n sina,
a = (N.A.) 2z/2. (All)
where n = diffraction index of object-sided medium (oil
or vacuum), M = enlargement factor and N.A. = nu-
merical aperture of the objective, e can be evaluated for References
any objective. In all cases << 1 holds, so that the 1. H. H. Hopkins, "The Frequency Response of a Defocused Optical
square root in Eq. (A2) can be expanded in a series and System," Proc. R. Soc. London Ser. A 231, 91 (1955).
approximated by the first two terms. Also, for small 2. P. A. Stokseth, "Properties of a Defocused Optical System," J. Opt.
image-side angles a' the approximation Soc. Am. 59, 1314 (1969).
3. W. K. Pratt, Digital Image Processing (Wiley, New York,
1978).
1 - cosa' - sin2 a' (A5)
2 4. J. W. Cooley and J. W. Tukey, "An Algorithm for Machine Com-
putation of Complex Fourier Series," Math. Computation 19, 297
is valid, so that Eq. (A2), can be rewritten:
(1965).
5. D. A. Agard and J. W. Sedat, "Threedimensional Architecture of
a-' N.A.2 Z r1-z ' (A6) a Polythene Nucleus," Nature London 302, 676 (1983).
6. D. Mathog, M. Hochstrasser, Y. Gruenbaum, H. Saumweber, and
which determines the relation between the image-sided J. Sedat, "Characteristic Folding Pattern of Polythene Chromo-
maximum phase error o' and the image-sided focal shift somes in Drosophila Salivary Gland Nuclei," Nature London 308,
z'. The relation between the image- and object-sided 414 (1984).
focal shift is given by7 7. H. G. Zimmer, Geometrische Optik (Springer, Berlin, 1967).

200 APPLIEDOPTICS / Vol. 24, No. 2 / 15 January 1985