Point Spread-Function, Line Spread-Function,

and Modulation Transfer Function

Tools for the Study of Imaging Systems!
KURT ROSSMANN, Ph.D.

N RECENT YEARS the transfer theory of

I linear, invariant communication sys- INPUT OUTPUT
tems has been of increasing usefulness in the : ..- SYSTEM <,

study of radiographic imaging systems I

I
(1-4). Indeed, its application is not con- I
I
fined to the imaging systems themselves I
I
but can be extended to the analysis of the :
I
TRANSFER I
I
entire radiological process involving ex- L CHARACTERISTICS -J

posing, imaging, and visual detection op- Fig. 1. Schematic representation of a transducer
erations (5). This analysis will eventually from the viewpoint of communication theory.
result in quantitative descriptions of the
inherent limitations of present radiological matical derivation is given in the AP-
processes and, hopefully, in the develop- PENDIX.
ment of improved processes yielding in- GENERAL DESCRIPTION OF BASIC CONCEPTS
creased diagnostic certainty.
At present the number of investigators The Point Spread-Function
working in this field is very small-s-too In the analysis of a physical system,
small, in fact, to insure a satisfactory rate methods of communication theory are used
of progress toward the important goal. to determine the performance of the sys-
This is partly due to the fact that the tem as a transducer in converting a system
physics and mathematics involved are input to an output. It is not the aim of
highly specialized and cannot readily be communication theory to investigate in
assimilated from the existing literature. detail the interior of a system but rather
Any author in this field is faced with the to characterize a system terminally by
dilemma of writing lucidly for readers hav- establishing a general dependence of the
ing diverse backgrounds of scientific train- output on the input. As indicated in
ing, since meaningful investigations in this Figure 1, the problem can be stated as
field cannot be carried out by physicists follows: Given a black box (the system),
alone but must be made in cooperation with determine its transfer characteristics so
radiologists who are the only ones, after all, that the output resulting from any con-
who fully appreciate the operational aspects ceivable input can be uniquely predict-
of the radiological process. In this co- ed.
operation each investigator will tend to The practical importance of knowing the
contribute most in the field for which he system transfer characteristics is obvious.
was trained. On the other hand, it is help- For example, if the system is a sound trans-
ful if all investigators develop a common mitter the "fidelity" of the output can be
language and an understanding of basic predicted; in the case of imaging systems
concepts. With this in mind, the present the image deterioration introduced for any
discussion of some important concepts of given object can be predetermined. The
optical communication theory is presented present discussion will be confined to
in nonmathematical form. A brief mathe- imaging systems.
1 From the Department of Radiology, The University of Chicago, Chicago, Ill., and the Argonne Cancer Research
Hospital (operated by The University of Chicago for the United States Atomic Energy Commission). Accepted
for publication in April 1969.
RADIOLOGY 93: 257-272, August 1969. jc

257
258 KURT ROSSMANN August 1969

(point source) radiating with unit intensity.

l~"limog, point
In a perfect imaging sytem the radiant
energy emanating from a point source in
the object plane would be concentrated

hObjectplane
Intensity
at a point in the image plane, the ideal
image point. In practical systems, how-
ever, optical imperfections result in a
"smearing-out" of the energy around the
ideal image point and, therefore, in un-
sharp imaging of the point source. The
point spread-function provides a measure
Fig. 2. The point spread-function is the image of a of this unsharpness. In Figure 2 the
unit intensity point source (arrow). a, nonisotropic unit point source is shown as an arrow
system; b, isotropic system.
of unit length standing on the object
In the general case, this system analysis is plane and the point spread-function
extremely complicated. It can be sim- as a "hump" on the image plane. As
plified, however, if the investigation is re- indicated schematically in Figure 2, A,
stricted to a particular type of system hav- the point spread-function is unsymmetric
ing the following properties: in general. For certain systems, however,
1. Linearity, which is distinguished by the point spread-function possesses rota-
the two basic characteristics: (a) the out- tional symmetry as shown in Figure 2, B.
put corresponding to a sum of inputs is Systems of this type are called isotropic.
equal to the sum of the outputs corre- The isotropic property results in a sim-
sponding to the inputs acting separately; plified description of the transfer character-
and (b) multiplication of the input by a istics of the system which will be discussed
constant multiplies the output by the same later. From the above, it is apparent
constant. that the point spread-function is, in fact, a
2. Invariance, which requires that the transfer characteristic of the system for a
image of a point retains its shape as the specific input in that it provides a unique
object point is moved in the object plane. relation between a unit point source input
This property is frequently called iso- and the corresponding output. It will now
planatism. Often this requirement is not be shown that the point spread-function is
fulfilled in the entire image plane. In of much broader significance for linear
that case the image plane can be decom- systems.
posed into isoplanatic patches over each of From the first property associated
which the shape of the point image does with system linearity described previ-
not vary appreciably. The shape of the ously it follows that, if an arbitrary
point image, however, does vary from number of unit point sources is located in
patch to patch. For the sake of clarity the the object plane, each of them will be
following discussion will be confined to imaged independently of the others as a
systems which are isoplanatic as a whole. point spread-function in the image plane.
Conclusions reached, however, will be The total image of all unit point sources
applicable within each isoplanatic patch in is then simply the sum of all corresponding
systems which are not isoplanatic as a point spread-functions over the image
whole. plane. This is known as the superposition
The isoplanatic property enables us to principle of linear imaging.
define a unique characteristic of the imag- From the second property of linear sys-
ing system which is called the point tems it follows that, if the intensity of a unit
spread-function of the system and is de- point source is multiplied by some constant,
fined as the radiation intensity distribution then the corresponding point spread-func-
in the image of an infinitely small aperture tion will be multiplied by the same constant
Vol. 93 POINT SPREAD-FuNCTION, LINE SPREAD-FuNCTION, AND MTF 259

to yield the image of the point source of Intensity Intensity

nonunit intensity. This property in con-

junction with the superposition principle
leads to the conclusion that, if the input
System
consists of a field of point sources of
arbitrary intensity, the output or the total
image of the field of point sources is simply Image plane

the sum of the corresponding point spread- Fig. 3. The image of two point sources formed by a
functions, each multiplied by an appro- linear, isotropic system (superposition principle).
priate constant to take account of the
intensity of the corresponding point source. Intensity

Figure 3 illustrates this phenomenon for 1---

two point sources.

It is conceptually not difficult to extend
the case of a finite number of point sources
to the practical case of a continuous, two- Object plane Image plane
dimensional object which can be considered Fig. 4. The line spread-function is the image of a
as an aggregate of an infinite number of unit intensity line source.
point sources of different intensities. The
image of each point in the object is the scanning with a small aperture exactly
point spread-function multiplied by an through the center of the distribution
appropriate intensity factor, and the total which causes alignment difficulties. These
image is the sum of all the point images. experimental problems which cause in-
Thus, from a knowledge of the input in- accurate measurements can be overcome
tensity distribution in the object and of by measuring another transfer character-
the system point spread-function the out- istic of the system from which the point
put intensity distribution in the image can spread-function can be calculated.
be determined. Therefore, the point
spread-function is a general transfer char- The Line Spread-Function
acteristic of linear, isoplanatic imaging For linear, isoplanatic imaging systems,
systems. The mathematical operation of a second transfer characteristic can be
multiplying each point in the object in- defined. This is the line spread-function
tensity distribution by the system point which represents the radiation intensity
spread-function and summing over the distribution in the image of an infinitely
entire object distribution is known as narrow and infinitely long slit (line source)
convolution of the input with the point of unit intensity. In a perfect imaging
spread-function. system the radiant energy emanating from
Before this convolution can be carried a line source in the object plane would be
out, the point spread-function of a given concentrated in a line in the image plane.
system needs to be measured by using a In practical systems, however, optical
small aperture as a source. Direct mea- imperfections result in a "smearing-out"
surement of the point spread-function, of the energy around the ideal line image
however, is difficult for two experimental and, therefore, in unsharp imaging of the
reasons. First, in order to approximate a line source. The line spread-function pro-
point source of radiation the aperture must vides a measure of this unsharpness. In
be made quite small relative to the size of Figure 4 the unit line source is shown as a
the point spread-function. Under prac- "knife edge" of unit height standing on
tical conditions this yields a very low input the object plane, and the line spread-func-
radiation intensity into the system. Sec- tion as a "welt" on the image plane. It
ond, measurement of the resulting intensity will be shown later that the line spread-
distribution in the image plane requires function is a system transfer characteristic
260 KURT ROSSMANN August 1969

Intensity
the imaging system is isotropic (7). In
that case the point spread-function is ro-
tationally symmetric, as indicated in Figure
~
System 2, B, and the shape of the line spread-
function is independent of the orientation
Object plane Image plane of the line source in the object plane and
Fig. 5. The image of two line sources formed by a
is also symmetric. Thus, if the system is
linear system (superposition principle). isotropic, one measurement of the line
spread-function suffices for the calcula-
in that it provides a unique relation be- tion of the point spread-function.
tween a certain class of arbitrary inputs To summarize, the line spread-function
and the corresponding outputs. serves as an experimentally accurate tool
In practice the line spread-function is for determining the point spread-function
measured by approximating a line source which is a system transfer characteristic
with a slit which is narrow and long relative for the most general case of two-dimen-
to the size of the point spread-function, sional radiation intensity distributions in
and by scanning the resulting output (the the object plane. In addition, it can be
slit image) with a narrow slit. This experi- shown that the line spread-function is a
mental technic eliminates both difficulties system transfer characteristic for the
associated with the direct measurement of special case of one-dimensional inputs.
the point spread-function. Determining The reasoning is analogous to the case of
the point spread-function from the mea- the point spread-function in the foregoing
sured line spread-function, however, is no section and will, therefore, be presented
simple matter in general. This is due to only in its essentials.
the fact that the line spread-function is a As discussed previously, the point spread-
one-dimensional function obtained from function is a unique characteristic of iso-
a rectilinear scan of a one-dimensional in- planatic imaging systems. Therefore, the
tensity distribution, whereas the point line spread-function is also a unique system
spread-function is two-dimensional. This characteristic for anyone orientation of the
can be explained by noting the relationship line source relative to nonisotropic systems
between the two functions. or for any orientation relative to isotropic
It can be shown mathematically (see systems. If the system is also linear, an
APPENDIX) that the direct measurement of input consisting of a field of line sources of
the line spread-function described above arbitrary intensi ty will result in an output
is equivalent to scanning the point which is the sum of the corresponding line
spread-function with a slit which is spread-functions, each multiplied by an
narrow and long relative to the size of the appropriate intensity factor. Figure 5
point spread-function. Since the point illustrates this for two line sources. If the
spread-function is often unsymmetric, as input is a continuous object over which the
indicated in Figure 2, A, the shape of the radiation intensity varies in one dimension
line spread-function depends on the direc- only, such as a straight-edge or a bar pat-
tion in which the point spread-function tern, the object can be considered as an
is scanned. For the calculation of the aggregate of an infinite number of line
point spread-function the line spread- sources of different intensities. The corre-
functions corresponding to all possible sponding output is calculated by multi-
orientations of the scanning slit must be plying each line source in the object in-
known (6). In terms of the directly tensity distribution by the system line
measured line spread-function this means spread-function and summing over the
that the line source must be placed in all entire object distribution. This one-di-
possible orientations in the object plane. mensional convolution operation is illus-
Matters are simplified considerably when trated in Figure 6. Therefore, the line
Vol. 93 POINT SPREAD-FUNCTION, LINE SPREAD-FuNCTION, AND J\;ITF 261

viElemental line sources

/Object
~""i~ distribution
I
1
I
I
I
Fig. 6. Schematic illustration 1
of one-dimensional convolution. Object plane : Distance
I
Only a few of an infinite number I
of elemental line images are Intensity I
I
shown. I
I
I
I

! /Object intensity
i times line spread
!
I
function
I

Image plane Distance

spread-function is a transfer characteristic cussed. It was seen that by means of the

of linear, isoplanatic imaging systems for system point spread-function, the output
the special case of one-dimensional inputs. resulting from an arbitrary, two-dimen-
It must be emphasized that the line sional input can be predicted. Similarly,
spread-function does not serve as a magical the system line spread-function can be used
shortcut to reduce a two-dimensional directly to predict the output correspond-
transfer problem to a one-dimensional one. ing to an arbitrary, one-dimensional input.
To describe the transfer of two-dimensional The great utility of these concepts derives
inputs, the point spread-function is needed, from the fact that the study of the trans-
and the line spread-function is merely an mission of complex object intensity dis-
accurate experimental tool to determine the tributions is reduced to the study of the
point spread-function. The line spread- transmission of very simple intensity dis-
function leads to a simplification of the tributions-namely, a point source or a line
overall problem only in the case of one- source. Once the relatively uncomplicated
dimensional inputs. Even in this case, the experiments to measure the transmission
two-dimensional character of the imaging of these simple object distributions have
system cannot be ignored entirely when the been carried out on a given system, the
system is nonisotropic. The slit which is transmission of any conceivable object
used to measure the line spread-function distribution can be calculated.
must be oriented in the same direction rela- In this context the study of the trans-
tive to the imaging system in which the mission of a third simple object intensity
direction of constant intensity in the one- distribution in which the intensity varies
dimensional object is criented. Only in sinusoidally with distance in the object
the study of isotropic systems can the plane is of particular usefulness. The solid
orientation of the object relative to the curves in Figure 7 depict two such distribu-
system be ignored. tions having different spatial frequencies
measured in cycles/rum. The term spatial
The Modulation Transfer Function (MTF) frequency does not imply a vibration or
In the foregoing sections two methods of change in time of the intensity distribu-
describing the optical transfer character- tion. The distribution is considered sta-
istics of imaging systems have been dis- tionary in time and space just as the line
262 KURT ROSSMANN August 1969

Intensity plane, and the modulation and phase

shift of each is measured in the image plane.
The modulations and phase shifts in the
image will vary with spatial frequency.
u,D;D.. D\ The ratio of the output modulation to the
"V'IU}V}" input modulation together with the phase
shift expressed as a function of spatial fre-
quency is called the optical transfer func-
tion of the system (8). The ratio of the
Distance output modulation to the input modulation
Fig. 7. Sinusoidal intensity distributions in space. alone, expressed as a function of spatial
Solid lines, input distributions of different spatial frequency, is called the modulation transfer
frequency; dotted lines, output from a linear non-
isotropic system. ' function of the system. In mathematical
terms the modulation transfer function is
pattern in a resolution chart which is de- the absolute value of the optical transfer
scribed as having a certain number of function. For a complete description of
lines/mm." When such a sinusoidal in- the transfer of sinusoidal inputs through
tensity distribution in the object plane is nonisotropic systems the optical transfer
imaged by means of a linear, isoplanatic, function is required since a phase shift
nonisotropic system of unit magnification, occurs in these systems. In isotropic sys-
the intensity distribution in the image plane tems the phase shift is zero, so that the
will also be sinusoidal with the same spatial modulation transfer function completely
frequency as the object distribution. The describes the transfer of sinusoidal inputs.
optical imperfections of the system, how- For the sake of simplicity the following
ever, will cause the amplitude of the image discussion will be confined primarily to the
distribution to be reduced, and the dis- modulation transfer function.
tribution as a whole will be shifted laterally The mere fact that the modulation trans-
relative to the object distribution as shown fer function provides a description of the
by the broken curves in Figure 7. The imaging of sinusoidal intensity distributions
reduced amplitude indicates a loss of resolu- is not sufficient reason for introducing it,
tion in the system, and the lateral shift is since the same can be done by convolving
described as a phase shift. I t is customary the sinusoidal distribution with the system
to characterize the sinusoidal distribution line spread-function. The great signifi-
in terms of its modulation rather than its cance of the modulation transfer function
amplitude. Modulation is defined as the lies in the fundamentally different manner
ratio of the amplitude to the average value in which it describes the transfer of
of the distribution. Note that the average sinusoidal inputs. It will be recalled that
value of a sinusoidal distribution of radia- the calculation of the output from the input
tion intensity cannot be smaller than the by means of the point or line spread-func-
amplitude of the distribution since nega- tions proceeds from a point-by-point
tive values of intensity are physically knowledge of the intensity distribution in the
meaningless. Therefore, the modulation object plane. Convolution of the object
cannot be greater than unity. distribution with the point or line spread-
In practice, several sinusoidal intensity function, which is expressed in terms of
distributions having different spatial fre- distance in the image plane, results in a
quencies but identical amplitudes and point-by-point description of the intensity
modulations are introduced in the object distribution in the output plane. In other
words, point and line spread-function are
2 In the analysis of systems for the imaging of mov- transfer characteristics of the system in the
ing phenomena, e.g., fluoroscopic systems, an additional spatial domain. Calculation of the output
temporal variation of the sinusoidal input needs to be
introduced. This case will not be discussed here. from the input sinusoidal distribution by
Vol. 93 POINT SPREAD-FuNCTION, LINE SPREAD-FUNCTION, AND MTF 263

Intensity

Amplitude

Fourier)

transformation

(0) (b)
Distance Spatial frequency
Fig. 8. Two equivalent methods of describing a signal: a, x-ray pattern of a
blood vessel; b, corresponding amplitude spectrum.

means of the modulation transfer function, cycles/rum as described above. Figure 8

on the other hand, proceeds from a knowl- shows an example of a signal and its ampli-
edge of the modulation and the spatial tude spectrum. This particular signal is
frequency of the input. Multiplication of the input to a radiographic imaging system
the input modulation by the modulation (5). In principle, describing the input as
transfer function, which is expressed in an intensity distribution in the spatial
terms of spatial frequency, results in the domain and as an amplitude distribution
modulation of the output having the same in the frequency domain is analogous to
spatial frequency. Thus, the modulation describing it in two different languages.
transfer function describes the transfer of The two descriptions are equally valid and
sinusoidal inputs in the spatial frequency comprehensive," and the Fourier transform
domain. Note that the mathematically is the means for translating from one
complicated convolution operation in the language to the other. For example, in
spatial domain is replaced by simple multi- Figure 8 the width of the input in the
plication in the spatial frequency domain. spatial domain translates into the magni-
(Compare equations (2) and (5) in the tude of the amplitudes at high frequencies
APPENDIX.) It will now be shown that the or, simply, the high-frequency content
modulation transfer function also provides in the frequency domain. Similarly, the
a unique relation between arbitrary, study of the transfer of intensity distribu-
not necessarily sinusoidal inputs and the tions in the spatial domain becomes a study
corresponding outputs and is therefore a of the transfer of amplitude spectra in the
general system transfer characteristic. frequency domain. Since the modulation
It is well known (9) that most non- transfer function describes the transfer of
periodic variations of a quantity in time or sinusoidal inputs in the frequency domain,
in space (signals, inputs, etc.) can be repre- it also describes the transfer of amplitude
sented as a sum of an infinite number of spectra. Therefore, the modulation trans-
sinusoidal component signals of different fer function is a general transfer character-
amplitudes and frequencies. The mathe-
3 In this discussion the amplitude spectrum is defined
matical process for accomplishing this as the absolute value of the complex Fourier spectrum,
harmonic analysis is called Fourier trans- which is in keeping with mathematical and engineering
formation. A plot of the amplitudes of the terminology. It is not to be confused with the Fourier
spectrum of an input amplitude distribution which is
component signals as a function of fre- often called an amplitude spectrum in optics.
4 In general the description in the frequency domain
quency is known as the amplitude spec- must also include the phases of the sinusoidal compo-
trum" of the original signal. In the tem- nents. Since the present discussion is concerned mainly
poral domain, the frequency is expressed in with the modulation transfer function and not with the
optical transfer function, this more complicated aspect
cycles/sec. and in the spatial domain in of the theory will not be included.
264 KURT ROSSMANN August 1969

SPATIAL DOMAIN the complete, two-dimensional modulation

transfer function. When the system is
OBJECT SYSTEM IMAGE isotropic, the modulation transfer function
FOURIER Intensity (Transducer) i Intensity is rotationally symmetric and the orienta-
TRANSFORM distribution i
: distribution
i
tion of the sinusoidal input in the object
Point spread-
function
plane is immaterial. In this case the
Line spread-
_____
J
modulation transfer function is usually
function (Convolution) plotted in one dimension only.
It is of great experimental significance
for situations in which it is difficult to
generate sinusoidal inputs, e.g., x-ray in-
FREQUENCY DOMAIN
puts, that the modulation transfer func-
OBJECT, SYSTEM IMAGE,
tion along a certain direction can be cal-
I ~ (Filter) ~ culated by one-dimensional Fourier trans-
Amplitude i i Amplitude
spectrum : -- i spectrum formation from the line spread-function
l_ Modulation transfer_J measured by means of a line source oriented
function (Multiplication) perpendicular to that direction (3). In
Fig. 9. Summary of concepts used in optical com- the case of isotropic systems, the orienta-
munication theory. tion of the line source is immaterial, and the
general modulation transfer characteristics
istic of linear, isoplanatic, isotropic systems of the system can be calculated simply
in the frequency domain, linking arbitrary from the line spread-function by one-
inputs with the corresponding outputs. dimensional Fourier transformation. As
By the same reasoning for nonisotropic in the spatial domain, this apparent reduc-
systems the optical transfer function is a tion of a two-dimensional problem to a one-
general transfer characteristic. Thus, just dimensional problem should not be mis-
as in the case of the point and line spread- interpreted. To analyze the transfer of
function, the introduction of the modula- two-dimensional inputs which are described
tion transfer function has reduced the by two-dimensional amplitude spectra in
study of the transmission of complicated the frequency domain, the two-dimensional
inputs to the study of the transmission of modulation transfer function is needed.
very simple inputs-namely, sinusoidal Only in the case of one-dimensional inputs
intensity distributions. In keeping with is the one-dimensional modulation transfer
the language of the frequency domain, function sufficient to relate the output to
imaging systems are often called filters of the input.
spatial frequencies (see APPENDIX). Figure 9 broadly summarizes the trans-
To complete the equivalence of the fer characteristics of imaging systems in the
spatial and frequency domains, the counter- spatial domain and in the spatial frequency
part of the modulation transfer function in domain which have been discussed. It
the spatial domain is the point spread-func- remains to be shown in what manner the
tion, since one can be calculated from the description in the frequency domain aids
other by two-dimensional Fourier trans- in the solution of the transfer problem. As
formation. Thus, the modulation transfer noted previously, calculation of the output
function is two-dimensional in the fre- corresponding to a given input in the spatial
quency domain and is rotationally un- domain requires convolution of the input
symmetric for nonisotropic systems. This with the point spread-function or the
means that, if one-dimensional sinusoidal line spread-function of the system. In the
inputs are used to measure the modulation frequency domain the output is calculated
transfer characteristics of non isotropic sys- simply by multiplying the input amplitude
tems, they must be placed in all possible spectrum by the modulation transfer func-
orientations in the object plane to obtain tion of the system. This reduction in
Vol. 93 POINT SPREAD-FuNCTION, LINE SPREAD-FUNCTION, AND MTF 265

mathematical complexity is especially use- date the whole frequency band into which
ful for the analysis of cascaded systems con- the pulse can be resolved. Otherwise, dis-
sisting of several linear, isoplanatic com- tortion will result. Similarly, the optical
ponents in series so that the output from degradation of an image can be related to
one component is the input to the next the modulation transfer function of an
component. In the spatial domain, cal- imaging system.
culation of the overall transfer character-
istics of such systems from the component RADIOGRAPHIC APPLICATION

transfer characteristics would require a In order to relate the concepts discussed

complicated multiple convolution. On the above to practice, their application to
other hand, the total modulation transfer radiographic screen-film systems will be
function of the complex system is simply described. The details of the experi-
the product of all component modulation mental procedure have been discussed in
transfer functions. This makes it easy, for previous publications (3, 11) and will only
example, to determine the weakest link be summarized here.
in the chain of imaging components com- The most commonly used screen-film
prising a composite system. system consists of a pair of fluorescent
Furthermore, working in the frequency screens in intimate contact with both sides
domain makes it possible to study the of a radiographic film which has photo-
combined effects on the image of the optical graphic emulsion coated on both sides of a
system characteristics and of noise, such as support. The thickness of the phosphor
electronic system noise in image intensifier- layer on each screen is about 100 M, and
television chains, grain noise in films, and each emulsion is about 10 J1 thick. The
input quantum noise in all radiographic thickness of the support is about 180 J1.
imaging systems. Noise is best described Before the above concepts of communica-
quantitatively by means of the Wiener tion theory are applied to this system, it
spectrum which is derived from Fourier needs to be established if the system satis-
analysis of the random noise pattern. The fies the conditions of linearity and iso-
Wiener spectrum indicates the spatial planatism.
frequency content of noise just as the The exposure response of the screen-film
amplitude spectrum is used to describe the system is nonlinear when it is expressed in
spatial frequency content of nonrandom terms of photographic density by means of
intensity distributions. For example, the the characteristic curve of the film. It has
modulation transfer function in con- been determined experimentally, however,
junction with the noise spectrum is useful that the system is linear if the output is
in predicting the transmission of input expressed in terms of effective exposure or
noise through the system. illuminance in the film. From two slit
Besides the two specific advantages of images whose exposures differed by a factor
the frequency domain representation cited of four, it was found that the ratio of the
above, treating optical imaging systems as two effective illuminance distributions in
filters of spatial frequencies brings to bear the film was equal to the measured x-ray
on the optical transfer problem a great body exposure ratio and independent of the
of knowledge derived from electrical com- distance from the center of the line image.
munications theory (10). For example, a Thus, the screen-film system can be "linear-
close analogy exists between the design of ized" by using as the output the effective
an optical system and the construction of a illuminance in the emulsions. This pro-
device to amplify an electrical pulse. De- cedure is commonly applied in the analysis
pending on the shape of the pulse which of photographic films.
determines its frequency content, the The isoplanatic condition requires that
amplifier must have a frequency response the point spread-function is independent of
sufficiently wide (band width) to accornmo- the angle at which x rays are incident on
266 KURT ROSSMANN August 1969

mediately adjacent to the intensifying

screen nearest the x-ray tube, and the input
as the x-ray intensity pattern in this plane.
Geometrical unsharpness and enlargement
effects are thereby separated from the
characteristics of the screen-film system
and are described by separate transfer
characteristics of the exposing process (12).
The line spread-function of screen-film
Fig. 10. Radiographs of a 1O-,u slit on Royal Blue systems is then measured by means of a 10
film. A. Without intensifying screens. B. With Mm slit formed by platinum jaws, 2000 Mm
medium-speed screens. C. With fast screens.
thick, which is mounted in the aluminum
the screen-film system. This is not the front of a vacuum-exposure holder. The
case, in general, since oblique incidence screen-film system to be in vestigated is
will result in an increase of the effective placed inside the holder in intimate con-
thickness of the screens and in lateral dis- tact with the slit and exposed to x rays on
placement of images. The influence of an optical-bench arrangement. The finite
oblique incidence on the point spread- width of the slit is negligible relative to the
function, however, is estimated to be unsharpness of present screen-film systems,
negligible for practical exposure setups in- so that the slit is effectively a line source of
volving focal-spot-screen distances which radiation. Exposed and processed slit
are very large relative to the thickness of images similar to those shown in Figure 10
the system for films of normal size. In are traced on a microdensitometer (13), and
any case, the isoplanatic condition will the line spread-function is calculated by
certainly be satisfied sufficiently well if means of the micro densitometer calibra-
only objects near the perpendicular to the tion curve and the characteristic curve of
screen-film system are considered. the film. It should be noted that this
In additon to being linear and isoplanatic, method of measurement results in a line
screen-film systems can also be expected spread-function whose physical meaning is
to be isotropic. This is known to be the somewhat different from that used in
case for photographic emulsions and should optics. Since the double-coated film sand-
also be true for intensifying screens if the wich is quite thick and since the total slit
screens are uniform, since it is difficult to image is composed of partial images in the
see how a uniform screen could introduce two emulsions which are separated by the
directional effects. Experimental results film base, a well-defined image plane does
have confirmed this expectation. not exist within the system. In fact, the
In applying linear communication theory only image plane which can be defined
to radiographic screen-film systems in- rigorously is the final image plane in the
volving double-coated films, it should be microdensitometer during the scanning
noted that the process of image formation operation. Therefore, scanning of the slit
is quite different from that usually en- image and subsequent application of the
countered in optics. The radiographic conventionally measured characteristic
image is essentially a shadowgram of a curve of the film do not yield the corre-
three-dimensional structure formed by sponding exposure or illuminance distribu-
penetrating radiation. The source of radi- tion in an image plane, but rather what
ation is of finite size and the x-ray beam is might be looked upon as an "effective" ex-
diverging. As a result, geometrical un- posure or illuminance distribution in the
sharpness and enlargement exist in the film. It has been shown experimentally
input to the imaging system. Under (14) that it is valid to call this effective
these conditions it is convenient to define distribution a line spread-function of the
the object plane as a plane in space im- screen-film system.
 1.0

wO.8
o
Z
<1
z:
~0.6
...J
:::!
w
~ 0.4
~
...J
W
a: 0.2

250 200 150 100 50 0 50 100 150 200 250

DISTANCE (microns)

Fig. 11. Normalized line spread-function of a medium-speed screen-film

system.

1.0
w
u
~0.8
z
~
:::>
...J
=:!0.6
w
>
ti...JOA
w
a:

0.2

250 200 150 100 50 0 50 100 150 200 250

RADIUS (microns)

Fig. 12. Normalized point spread-function calculated from Figure 11.

1.0
0: .9
0
I-
0 .8
~
0: .7
lJJ
LL
(f) .6
z
«
0: .5
I-
Z .4
0
~...J .3
::J
0 .2
0
~ .1

2 3 4 5 6 7 8 9 10 II 12 13 14
SPATIAL FREQUENCY (Cycles/rnrn)
Fig. 13. Modulation transfer function calculated from Figure 11.
267
268 KURT ROSSMANN August 1969

From the measured line spread-func- REFERENCES

tion, one can calculate the point spread- 1. DOl, K., AND SAYANAGI, K.: Optical Transfer
function using a method described by Function in Radiography (I): X-ray Fluorescent
Screen. Oyo Buturi 33: 721-726, 1966.
Marchand (7), and the modulation transfer 2. MORGAN, R. H., BATES, L. M., GOPALA RAO,
function by Fourier transformation. Re- U. V., AND MARINARO, A.: The Frequency Response
Characteristics of X-ray Films and Screens. Am. J.
sults for a medium-speed screen-film sys- Roentgenol. 92: 426-440, August 1964.
tem are shown in Figures 11-13. Applica- 3. ROSSMANN, K.: Measurement of the Modula-
tion Transfer Function of Radiographic Systems Con-
tion of these transfer characteristics to the taining Fluorescent Screens. Physics in Med. & Biol.
calculation of outputs from certain inputs 9: 551-557, October 1964.
4. STAHNKE, 1., AND HEINRICH, H.: The Modula-
has been discussed elsewhere (5). tion Transfer Function of an Electron-optical X-ray
The analysis of radiographic imaging Image Intensifier Tube. Optik 23: 251-267,1966.
5. ROSSMANN, K.: Spatial Frequency Spectrum:
systems by means of such communication Means for Studying Quality of Radiographic Imaging
theory methods can provide the following Systems. Radiology 90: 1-13, January 1968.
6. MARCHAND, E. W.: From Line to Point Spread
valuable results: (a) objective measure- Function: The General Case. J. Optic. Soc. America
ment of the optical system performance; 55: 352-354, April 1965.
7. MARCHAND, E. W.: Derivation of the Point
(b) determination of the optical degrada- Spread Function from the Line Spread Function. J.
tion of diagnostically important detail in Optic. Soc. America 54: 915-919, July 1964.
8. INGELSTAM, E.: Nomenclature for Fourier
the image; and (c) prediction of optical Transformers of Spread Functions. J. Optic. Soc.
system parameters for the design and use America 51: 1441, December 1961.
9. BRACEWELL, R.: The Fourier Transform and
of systems providing optimum visibility of Its Applications. New York, McGraw-Hill, 1965.
diagnostically important detail. 10. HANCOCK, J. C.: An Introduction to the Prin-
ciples of Communication Theory. New York, Me-
The same concepts can be applied to the Graw-Hill, 1961.
analysis of other phenomena affecting 11. ROSSMANN, K., LUBBERTS, G., AND CLEARE,
H. M.: Measurement of the Line Spread-Function of
radiographic-image quality, such as geo- Radiographic Systems Containing Fluorescent Screens.
metrical and motion unsharpness and the J. Optic. Soc. America 54: 187-190, February 1964.
12. LUBBERTS, G., AND ROSSMANN, K.: Modula-
recording of quantum noise. tion Transfer Function Associated with Geometrical
Unsharpness in Medical Radiography. Physics in Med.
SUMMARY & BioI. 12: 65-77, January 1967.
13. ALTMAN, J. H., AND STULTZ, K. F.: Micro-
Concepts of communication theory are densitometer for Photographic Research. Rev. Sci-
ent. Instruments 27: 1033-1036, December 1956.
discussed which can be used to determine 14. ROSSMANN, K., AND SANDERSON, G.: Validity
the optical performance of imaging systems of the Modulation Transfer Function of Radiographic
Screen-Film Systems Measured by the Slit Method.
in converting a system input to an output. Physics in Med. & BioI. 13: 259-268, April 1968.
Three transfer characteristics of linear,
APPENDIX
invariant imaging systems are described in
a nonmathematical manner: the point Discussion of Spatial Frequency Concept
spread-function and line spread-function The description of an x-ray pattern in space as a
in the spatial domain, and the modulation sum of many sinusoidal components of different
amplitudes and spatial frequencies can be illustrated
transfer function in the spatial frequency best by means of a specific example.
domain. The physical meaning of these Let us assume that the x-ray pattern of interest
concepts and the relationships between is the square wave, !(x), shown in Figure 14. By
them are explained in detail. A brief means of a mathematical technic called Fourier
mathematical derivation is given in the analysis, this function of distance can be written as a
sum of infinitely many sinusoidal functions,
APPENDIX.
The application of linear communication f (x) = ~ + 2 sin x + ~ sin 3 x + ~ sin 5 x + ....
theory to radiographic screen-film systems
is discussed. Measurement of the line The first term in this Fourier series representation
spread-function of these systems is ex- of the function is a constant and is often called the
DC-term. Subsequent terms are sinusoidal and
plained, and transfer characteristics are each is characterized by an amplitude A and a fre-
given for a medium-speed screen-film sys- quency II in keeping with the conventional nomen-
tem. clature for sinusoidal distributions (Fig. 15),
Vol. 93 POINT SPREAD-FuNCTION, LINE SPREAD-FuNCTION, AND MTF 269

f(x) y 1 cycle
I 11 I
I I
I I
I I
I I xlrnm)
I I
I I
I I
-Tf 0 rr xtmrn) Frequency = 1I cycles/mm
Fig. 15. Conventional nomenclature for sinusoidal
Fig. 14. Hypothetical x-ray intensity distribution in distributions.
space: f(x) = 0 for -7r < X < 0; f(x) = 7r for 0
< x < ar,

y = A sin 27rllx •

Since the distributions are sinusoidal in space, their 2 2

frequencies are expressed in cycles/rnm and are w
Q
called spatial frequencies. The plot of amplitude ::>
versus spatial frequency shown in Figure 16 is called t-
the spatial frequency spectrum or amplitude spec- :J
trum of the x-ray pattern. It expresses the x-ray a..
~
pattern in a form which is advantageous for studying <t: 2
the manner in which it is imaged. '!
The great significance of the amplitude spectrum
is that it tells us exactly and uniquely what sinus-
oidal distributions need to be added together to syn-
I
!iT In in
thesize the original x-ray pattern. In other words, FREQUENCY (cyctes/mrn)
the amplitude spectrum describes the spatial fre-
Fig. 16. Amplitude spectrum of intensity distribu-
quency content of the x-ray pattern. If any of the tion in Figure 14.
component sinusoidal distributions are not added,
the original x-ray pattern will not be obtained.
This is illustrated in Figure 17, which shows the
original x-ray distribution and the results of plotting
the sums of an increasing number of terms in the
Fourier series up to the fourth term. It is seen that
the original distribution is approximated more closely
the more terms are added. Note that one effect of
not adding all sinusoidal distributions which are
contained in the original pattern is a rounding of
the corners.
To translate these facts into the language of image
formation, we look upon the amplitude spectrum of
the x-ray pattern as the object for an imaging sys- Fig. 17. Synthesis of intensity distribution of Figure
tem. If the system is able to transmit the entire 14 by adding sinusoidal distributions given by spectrum
frequency content of the object, all component of Figure 16. Curve 1: y = 7r/2; Curve 2: y =
sinusoidal distributions necessary to synthesize the 7r/2+2 sin x; Curve 3: y = 7r/2+2 sin x+2/3 sin 3x;
Curve 4: y = 7r/2+2 sin x+2/3 sin 3x+2/5 sin 5x.
object will also be present in the image. This
means that the image will look like the object or
that the system is able to image the object perfectly. rounded corners resulting from the limited fre-
If, on the other hand, the system is not able to quency content of the image will give the impression
transmit the entire frequency content of the object, of unsharp imaging of the object.
i.e., if it acts like a :filter of spatial frequencies having It follows, therefore, that a knowledge of the filter-
insufficient band width, then the absence of these ing characteristics of an imaging system would
frequencies from the image will cause the image to enable us to predict how the object is transmitted
appear different from the object. Specifically, if by the system and what the appearance of the re-
the system transmits only sinusoidal distributions sulting image will be. In other words, the filter
at the low frequency end of the object spectrum, function of the system is a measure of the optical
i.e., if it is a low-pass filter, the effect on the image quality of the system. This filter function is called
will be similar to that shown in Figure 17. The the system modulation transfer function (1~1TF).
270 KURT ROSSMANN August 1969

Fig. 18. Top. Part of a sinusoidal transmittance test object for the direct measurement of modulation transfer
functions in optics. Bottom. Un sharp image of the test object.

The direct measurement of the MTF can best be Mathematical Derivation

illustrated by using an example from optics. Fig- Definitions: The one-dimensional Fourier trans-
ure 18, A shows a reproduction of a part of a trans- form of a functionf(x) is defined to be
parent piece of film which can be used as a test ob-
ject for the measurement of the MTF of lens-film
combinations, for example. The photographic den-
F(II) = f: co f(x)e-21ripxdx.
sity within each small rectangular test patch on the
film is such that the light intensity transmitted by It exists if
the film varies sinusoidally with distance along one
dimension of the patch. The spatial frequency of f:co If(x)ldx < co andf(x) has a finite number of
the sinusoidal light intensity distribution is different discontinuities.
from patch to patch. On this particular test object,
it ranges from 3/8 to 6 cycles/rum. When these The two-dimensional Fourier transform of a func-
sinusoidal light intensity distributions are used tion f(x,y) is defined to be
as objects for an unsharp lens, i.e., if they are imaged
unsharply by a lens on a piece of film, the resulting
image will appear as in Figure 18, B. Note that the
F(IIx,II u ) = ff: co f(x,y)e- 21ri(pxx + vyy) dxdy.

lens has not transmitted the higher spatial fre- It exists if

quencies contained in the object. The light in-
tensity distributions in the object and in the image
can be drawn schematically as in Figure 19, A.
The object has equal amplitude at all spatial fre-
fJ: 00 If(x,y) Idxdy < 00 and f(x,y) has a finite num-
ber of discontinuities.
quencies. In the image, the amplitude decreases
with increasing frequency. A plot of the ratio of The one-dimensional delta function o(x) is defined
image amplitude to object amplitude versus the by
spatial frequency of the corresponding patch is the
modulation transfer function of the lens-film com-
bination, as indicated schematically in Figure
f: co o(x)dx = 1 and o(x) = 0 for x ::f: O.
19, B.
The MTF shows that the lens-film combination is a
The two-dimensional delta function o(x,y) is
defined by
low-pass spatial frequency filter. When the ampli-
tude spectrum of any arbitrary object is multiplied
by the MTF, the amplitude spectrum of the image
is obtained.
ff: co o(x,y)dxdy = 1 and o(x,y) = 0 for x 2 + y2 ::f: O.

In the case of x-ray imaging systems, the system

MTF is not usually measured directly in this man- The Point Spread-Function
ner, since it is difficult to generate a sinusoidal spatial Consider an imaging system which operates on an
distribution of x-ray intensity which can serve as a input i(~,'YJ) to produce an output o(x,y)
test object. The MTF is calculated from the mea-
sured line spread-function instead. L{ i(~,'YJ)} = o(x,y) ,
Vol. 93 POINT SPREAD-FuNCTION, LINE SPREAD-FUNCTION, AND MTF 271

13 1.0
Q)
:c
1\/\1 ~
o

01

M -
C
Imaging E 0.5
<t-
System Il
u,
~
~

J'J\J\ (0) TEST PATCH FREQUENCY (cycles/mml

Fig. 19. A. Light intensity distributions transmitted by three of the test patches in Figure 18 (schematic).
B. Modulation transfer function calculated from intensity distributions in Figure 18 (schematic).

where (~,1J) and (x,y) are rectilinear coordinates in cates that the coordinate system in the object plane
the object plane and in the image plane, respectively. has been rotated through an angle 8 relative to the
Assume that the system is linear, such that original coordinate system (~,1]). This rotation re-
sults in new coordinate systems Ce,1]') and (x',y').
L{ alil(~,1]) + a2i2(~,1])} = alol(x,y) + a202(x,y) According to equation (2), the resulting output
= a1L{i 1 U ,1]) } + becomes

Any two-dimensional input can be expressed as a

a~{ i 2(t 1]) }.
o(x',8) = II: 00 a(e,8)pSF(x'. - e, y' - 1]',8)ded1]'
superposition of two-dimensional delta functions or
point source inputs, = I_OOoo PSF(x',y' - 1]',8)d(y' - 1]') =

LSF(x',8)

which is the line spread-function of the system along

Then, a certain direction. For isotropic systems the direc-
tional effect does not exist. This definition indi-
cates that the line spread-function can be obtained
either by using a line source input or by scanning the
point spread-function with a slit.
To calculate outputs resulting from two-dimen-
because of the linearity condition.
sional inputs by means of equation (2), the point
If the system is isoplanatic, a point spread-function
spread-function is needed. In can be calculated from
(PSF) or an output resulting from a point source
the measured line spread-function (6, 7). In the case
input can be defined, whose shape is independent of
of one-dimensional inputs iCe,O), however, we obtain
the position of the point source in the object plane,
from equation (2)
L{ a(t1])} = PSF(x - ~,y - 1])
o(x',O)
=
Then, equation (1) becomes
L{a(~ - e,1] - 1]')}.
= If:oo i(e,O)PSF(x' - e, y' - 1]',O)ded1]'

o(x,y) = II_oooo i(~,1])PSF(x - ~,y - 'fJ)d~d1]. (2) ~ I_mm i«',9{I:m PSF(x' - e,y' -",.9)d"'}e
This is a two-dimensional convolution integral, = I:oo i(e,O)LSF(x' - r,O)de. (3)
which states that the point spread-function is a trans-
fer characteristic of the system.
This is a one-dimensional convolution integral,
The Line Spread-Function which states that the line spread-function is a system
transfer characteristic for one-dimensional inputs.
Experimentally it is more convenient to work with
line source inputs a(~). For nonisotropic systems,
the orientation of the line source in the object plane
The Modulation Transfer Function
will affect the resulting line image. To emphasize To express the system transfer characteristics in
this fact, the line source is written aCe,e) which indi- the spatial frequency domain, we take the two-di-
272 KURT ROSSMANN August 1969

mensional Fourier transform of both sides of equa- This can be written

tion (2),
O(vx"O) = I(vx"O) . H(vx"O),
II:oo o(x,y)e- 21ri(Jlxx + Jlyy)dxdy = which shows that the Fourier transform of the line
spread-function is the optical transfer function along

If.[ff. i(,,~)PSF (x - "y - ~)d,d~J X

a certain direction V x'. For isotropic systems, the
directional dependence vanishes.
The connection between the optical and modula-
e- 21ri(Jlxx + JlYY)dxdy = II: 00 i(~,71) X
tion transfer functions and the transfer of one-di-
mensional sinusoidal inputs can be derived by sub-
stituting
e- 2• i (,", + ",l [II:. PSF(x - "y -~) X i(e,O) = A + Ai cos 27rvee
in equation (3). Letting x' - e= the output
~) }'d~.
0-,
e- 2• i [ , , (% - ,l + ,,(y - ,ll d(x - ,)d(y - becomes

This can be written o(x',O) = A + Ai I: 00 LSF(o-,O) cos 27rve

O(vx,vy ) = I(vx,v y) X H(vx,v y ) (4) (x' - rJ)drJ,
where 0 and I are the complex spatial frequency spec- and expanding the integrand,
tra of the output and the input respectively and His
the optical transfer function of the system. This o(x',O) = A + A dHe(v~') cos 27rvex' +
equation indicates that H(vx,vy) is a system transfer H 8 ( v e ) sin 27rv~'x'], (6)
characteristic in the frequency domain. Note that
the optical transfer function is the two-dimensional where He and H 8 are the cosine and sine transforms
Fourier transform of the point spread-function. of the line spread-function respectively. This can
Taking absolute values of the functions in equation be written
(4), we obtain o(x',O) = A {I + A1 0 cos [27rvex' - <I>(ve)]}, (7)
\O(vx,vy)j = !I(vx,vY)1 . MTF(vx,v y) (5) where M o = A o/ A is the output modulation.
Letting cos <I>(ve) = He(vd/(H/' + H 2) 1/2 and
which is a relation between the output and input
amplitude spectra and the modulation transfer func- sin <I>(ve) = H e2 + H 2 h in equation (6),
8(vd/(H 8 r 8

tion of the system. The modulation transfer func- the output becomes
tion is a general system transfer characteristic for o(x',O) = A {I + J.11 i .
isotropic systems, since in that case H(vx,v y)is real.
To express the transfer of one-dimensional inputs (H e2 + H 2)1/2 . cos [27rvex' -
8 1>(ve)]}, (8)
in the frequency domain, we form the one-dimen-
sional Fourier transform of both sides of equation where 111 i = AdA is the input modulation. Com-
(3), paring equations (7) and (8), we obtain
u, +H 2) 1/2 = .Mi X MTF(V~'),
I_oo 00 o(x',O)e-21riJlx,X'dx' = II_oooo i(e,O)LSF (x' -
Mo = X (H e2 8

since (H c2 + H 8 2 r /2 is the absolute value of the op-

tical transfer function. This result states that the
e,O)e-21riJlx,X'dx'de = I:oo i(e,O)e-21riJlx,e X system acts as a filter of spatial frequencies, the MTF
being the filter function.

[I:. LSF(x' - ",8)e- 2'<' ,'(%' -<'l X

The phase shift 1>(ve) in the output is
<I>(ve) = arctan H 8/H e.

d(x' - n}". This equals zero when H.

isotropic systems.
= 0, which is the case for