Professional Documents
Culture Documents
A Representation For Mammographic Image Processing: Ralph Highnam, Michael Brady and Basil Shepstone
A Representation For Mammographic Image Processing: Ralph Highnam, Michael Brady and Basil Shepstone
c Oxford University Press
Departments of Engineering Science and Radiology, Oxford University, South Parks Road, Oxford
OX1 3PJ, UK
Abstract
Mammographic image analysis is typically performed using standard, general-purpose algo-
rithms. We note the dangers of this approach and show that an alternative physics-model-based
approach can be developed to calibrate the mammographic imaging process. This enables us
to obtain, at each pixel, a quantitative measure of the breast tissue. The measure we use is h int
and this represents the thickness of ‘interesting’ (non-fat) tissue between the pixel and the X-ray
source. The thicknesses over the image constitute what we term the h int representation, and it
can most usefully be regarded as a surface that conveys information about the anatomy of the
breast. The representation allows image enhancement through removing the effects of degrading
factors, and also effective image normalization since all changes in the image due to variations
in the imaging conditions have been removed. Furthermore, the h int representation gives us a
basis upon which to build object models and to reason about breast anatomy. We use this ability
to choose features that are robust to breast compression and variations in breast composition. In
this paper we describe the h int representation, show how it can be computed, and then illustrate
how it can be applied to a variety of mammographic image processing tasks. The breast thickness
turns out to be a key parameter in the computation of h int , but it is not normally recorded. We
show how the breast thickness can be estimated from an image, and examine the sensitivity of
h int to this estimate. We then show how we can simulate any projective X-ray examination and
can simulate the appearance of anatomical structures within the breast. We follow this with
a comparison between the h int representation and conventional representations with respect to
invariance to imaging conditions and the surrounding tissue. Initial results indicate that image
analysis is far more robust when specific consideration is taken of the imaging process and the
h int representation is used.
Keywords: image enhancement and restoration, image simulation, mammography, surface
representation
Received September 14, 1995; revised December 4, 1995; accepted December 11, 1995
1. INTRODUCTION die of breast cancer than any other form of cancer. About
one woman in 12 can expect to develop breast cancer during
X-ray mammography continues to be the best examination for her lifetime. It has been demonstrated that early detection
early detection of breast cancer in post-menopausal women, is greatly improves mortality rates, perhaps by as much as 25%
the basis for national screening programmes and is the subject as claimed in the Forrest Report (Forrest, 1986), which led
of the mathematical and computational modelling proposed to the UK breast cancer screening programme based on X-
here. Women under the age of 50 years are currently excluded ray mammography, though this figure is currently the topic
from screening because of the attenuation of X-rays by dense of some debate (see below). For this reason, mammographic
breast tissue before involution. The facts about the incidence examinations are nowadays performed on about 25 million
of breast cancer are well known: more women in the West women annually in the EC (of which, about 3 million are in
∗ Correspondingauthor
the UK), at a cost of about US $3 billion per year. This huge
(e-mail: rph@robots.ox.ac.uk) cost and the poor accuracy in diagnosis, 8–25% of cancers are
2 R. Highnam et al.
Figure 1. The Xmammo interface. A left–right breast pair is shown. The buttons on the left-hand side enable the user to select the original image,
the primaries (scatter removal), to simulate the familiar hotlight, etc. A hotlight is shown as the bright square and allows close examination of
the densities within the light, radiologists use it in practice to view the breast edge. The buttons are repeated on the right side. At the bottom,
there are simple ways to change the time of exposure, the film and tube types, and to reposition the AEC.
missed and 70–80% of open surgical biopsies turn out to be As an application of image processing, mammographic
benign (Shroff et al., 1991), have led to increased interest in images pose a tough challenge because they have poor
applying computer-aided techniques. signal-to-noise ratio, typically 5–6 dB, corresponding to
A recent study (Woodman et al., 1995) suggests that the a noise level of approximately 3–4 grey levels in intensity
rate of presentation of tumours that arise within the screening (Cerneaz, 1994; Cerneaz and Brady, 1995). This is largely
interval or ‘interval cancers’ (the interval is three years in the because the images exhibit complex textures, because of
UK) is much larger than was predicted in the Forrest Report. If scattered photon radiation (Highnam et al., 1994) and
these early results on interval cancers hold up to further study, because there is a compromise between radiation dose and
this will cause a sharp decrease in the claimed improvement image quality. Worse, abnormalities appear as quite subtle,
in mortality as a result of the current protocols for national irregular, often non-local differences in intensity. Moreover,
screening. In fact, the findings have already led inter alia the images are inevitably cluttered due to superimposition,
to pressure for a reduction in the screening interval and to the background varies greatly between different breasts, and
routine adoption of two-view screening (cranio-caudal and there is relatively weak control of the imaging acquisition.
45◦ medio-lateral). The massive increase in the number of In short, mammographic images strain techniques of image
mammograms that these changes to the screening programme processing to their current limits, and beyond.
would entail make the development of reliable and robust Unfortunately, while there have been many papers written
computer techniques vital. on the application of image processing to mammography,
Mammographic image processing 3
the vast majority of the work has been of limited scope and
incorporates only general non-mammography-specific image
processing considerations. The dangers are obvious: image
smoothing may make lesions easier to locate, but can remove
calcifications and spiculations; edge sharpening may appear
Figure 3. Depiction of the h int surface of a breast, after smoothing
to improve an image from the image processor’s perspective
with a Gaussian G(0, σ ) where in the case shown σ = 3. This is
but can transform a malignant lesion into one that appears to a implemented by a convolution of mask size 5 by 5.
radiologist to be benign. In addition, currently available image
processing systems use terms familiar to engineers but totally
unfamiliar to clinicians, and this leads to unreliability, a lack
of confidence in users (typically radiographers) and disuse. by breast tissue that gives mammograms their appearance of
Full literature reviews can be found in Highnam (1992) and being slightly out of focus. There are many techniques for
Cerneaz (1994). sharpening up an (mammogram) image, of which unsharp
Our general approach has been that in order to be reliable masking is the best known. For the effects of scatter
and predictable, image processing must be based on a model removal to be predictable and reliable it is important that
of how the image is formed (Highnam, 1991, 1994, 1995). For while sharpening up the image to remove scatter, one must
mammography, this means we must model the way that X-rays be careful not to introduce artefacts that could change the
pass through breast tissue, are absorbed, and scattered before diagnosis! Technically, this corresponds to choosing carefully
exposing the film. Our modelling uses the work of a number the convolution kernel used in the unsharp masking. We
of medical physicists, most notably Haus et al. (1977), Day have shown, for example, that the Gaussian kernel, widely
and Dance (1983), Dance and Day (1984), Johns and Yaffe used in unsharp masking, gives poor results. The kernel we
(1987), Carlsson et al. (1989) and Dance et al. (1992). We developed is based on a model of the scattering of the X-rays
have also worked very closely with radiographers and clinical and their passage through an anti-scatter grid before absorption
radiologists, at the Churchill Hospital, Oxford and the Institut by the intensifying screen. We have shown mathematically
Gustave Roussy, Paris. and clinically that our procedure significantly improves the
A clear example of the difference between our approach and image, and hence the diagnosis, without introducing clinically
conventional approaches is in our modelling of the imaging suggestive artefacts. Further image restoration and enhance-
process to enhance mammograms by removing the effect of ment algorithms have been developed and incorporated into
scatter (Highnam et al., 1994). It is the scattering of X-rays a software system (Xmammo) that is available under license
4 R. Highnam et al.
Figure 4. Depiction of the h int surface of a breast, following processing of a mammogram by the Xmammo package. The height corresponds
to the amount of non-fatty tissue. The surface has been smoothed for easier viewing. The smoothing kernel is a Gaussian as in the previous
figure, with σ = 5. A cyst is clearly visible at the bottom left as a significant hill on the landscape. In this way, anatomical features correspond
to topographic features of h int surfaces.
from Oxford University and which is currently being evaluated the breast tissue at each pixel. The breast measure we use is
in several hospitals. A typical window display of Xmammo is the amount of non-fat (‘interesting’) tissue between the pixel
shown in Figure 1. The interface and screen layout result from and the X-ray source. We use h int to represent the thickness
substantial and continuing interactions with radiographers at of interesting tissue above each pixel and call these values
the Churchill Hospital, Oxford. collectively the h int representation. As a typical example,
Modelling the imaging process and understanding the Figure 2 shows a mammogram with a mass clearly located
object being imaged allows us to perform effective image nor- to the left of the nipple. Figures 3 and 4 are two displays of
malization. A mammogram contains two sorts of information, the h int surface computed from the image shown in Figure 2.
namely image variations that are due to: (i) the anatomical The tumour is clearly visible as a prominent hill in the south
structure of the breast, including pathologies and (ii) the choice west portion of the surface map.
of imaging parameters that are particular to the examination. In this paper we first describe the h int representation
The former include fibrous tissue, milk ducts, blood vessels informally (section 2). Then in section 3 we show how it
and fatty tissue, as well as calcifications and masses. The latter can be computed. The breast thickness turns out to be a
include: the type of X-ray tube and film used; the position of key parameter in the estimation of h int , but is not normally
the breast in the machine; the compression of the breast and recorded. We show how the breast thickness can be estimated
the exposure time. It is the former information that is required from the image, and the sensitivity of h int to this estimate.
for early diagnosis by both radiologists and computer. After The major contributions of the paper are in sections 4–8 in
obtaining calibration data and using appropriate models, we which we illustrate how the h int surface representation can
have shown how to remove the imaging effects (ii) to enable be applied to a variety of mammographic image processing
a radiologist (or computer) to concentrate on (i). The way tasks. In particular, it can be used for a range of image
we do this is to take sufficient calibration data so that we restoration and enhancement operations, and can form the
may transform the image to attain a quantitative measure of basis for simulating the appearance of anatomical structures
Mammographic image processing 5
within the breast. We compare the h int representation and Table 1. X-ray linear attenuation coefficients for various breast tissue
conventional representations with respect to invariance to types reported by Johns and Yaffe (1987).
imaging conditions and surrounding tissue. Initial results
demonstrate that image analysis is far more robust when µ (cm−1 ) at energy (keV)
specific consideration is taken of the imaging process and the Tissue type 18 20 25
h int representation is used. We conclude by sketching further
possible applications of the h int representation, including Minimum 0.538 0.441 0.314
matching mammograms (and breast MRI volumes) and as the Fat Mean 0.558 0.456 0.322
Maximum 0.585 0.476 0.333
basis for a fresh approach to feature extraction.
3.1. Calibrating the mammographic process a more accurate assessment of radiation dose. It should be
3.1.1. System calibration noted that most mammography compression devices have a
The mammographic imaging process has several parts which slant of about 0.5 cm across the breast image. Our computer
might vary from day to day. In order to effect meaningful software takes this slant into consideration, but in this paper
image analysis by computer, it is necessary to know these we consider H constant.
variations in order to make the images conform to a standard.
To achieve this requires calibration data. In our work, we 3.2. Computing hint
calibrate the film–screen response, film processor and film Given a mammographic image, we seek to find the thicknesses
digitizer. We assume that the X-ray tube output spectrum is of interesting and fatty tissue between the X-ray source and
relatively stable but that the number of incident photons varies each pixel. To do this, we consider the energy imparted to
across the image (the anode heel effect). In order to calibrate the intensifying screen at each pixel since we can attain these
these parts of the system we collect the following data: from the pixel values in the image using the calibration data.
Let E ps (x, y) be the energy imparted to the screen in the area
• A step wedge film. We need a film with a lucite step corresponding to the pixel (x, y). E ps (x, y) contains both
wedge placed along the back of the film with a lucite scatter and primary components. The primary component
block placed over the automatic exposure control. This E p (x, y) is determined by subtracting a scatter estimate from
film allows us to calibrate the film–screen system and film the total energy imparted. The process of estimating the scatter
processing so that energy imparted to the intensifying component is explained fully in Highnam et al. (1994). We
screen can be related to film density. compare E p (x, y) to the primary energies that we expect to
• A ‘blank’ film. We need a film taken with a short find in practice in order to determine h int (x, y).
exposure time with no object (breast) present. The For a pixel with h int cm of interesting tissue and h fat cm of
exposure has to be short so that the film does not saturate. fatty tissue above the corresponding area of the intensifying
We find that an exposure of 0.04 s, at 100 mA and 28 kV screen, the total attenuation at any energy E is expected to be
produces a film that has film densities that vary between
1.8 and 2.6 (despite looking black). This film provides hµ(E, x, y) = h int (x, y)µint (E) + h fat (x, y)µfat (E)
us with information about the spatial variations of the = h int (x, y)(µint (E) − µfat (E)) + H µfat (E),
incident radiation intensity.
• The digitized image of the step wedge film. The film (1)
density on each step of the wedge is measured so that where we have substituted h fat (x, y) = H − h int (x, y). In this
once digitized, the relationship between pixel value in the case, the energy expected to be imparted to the intensifying
digital image and the film density in the corresponding screen by the primary photons is
area of the film is known.
Vtube
3.1.2. Image calibration E p (x, y) = φ(Vtube , x, y)Ap ts N0rel (E)E S(E)G(E)
0
As well as calibrating the system components we need to
×e−µluc (E)h plate e−hµ(E,x,y) dE (2)
know data specific for each mammographic examination. In
particular, we require: where φ is the photon flux for an X-ray tube voltage of Vtube ,
• The tube voltage (Vtube kV); this varies across the image due to the anode heel effect; Ap is
• The tube current (Itube mA); the pixel area; ts is the time of exposure; N0rel (E) is the relative
• The time of exposure (ts s); number of photons at energy E; S(E) is the absorption ratio
• The breast thickness (H cm). of the screen to primary photons of energy E; G(E) is the
transmission ratio of the grid for primary photons of energy
Most of this information is readily available, but measuring E; µluc (E) is the linear attenuation coefficient of lucite at
the breast thickness H is currently awkward since the radiog- energy E and h plate is the thickness of the lucite compression
rapher has to measure it using a ruler; though newer machines plate.
are incorporating automatic measurement of breast thickness. Note that after substituting Equation (1) into Equation (2)
In both cases, the values of H have been shown to be inaccurate the only unknown is h int (x, y). We equate the primary energy
(Burch and Law, 1995). After explaining how the h int values found in the practical case with the theoretical value and
are generated, we detail ways of improving the accuracy of solve the resulting nonlinear equation to determine h int (x, y).
H using the image itself and discuss the effects of errors in Our experiments in over 100 clinical tests, carried out on
H on the values of h int . An improved estimate of H allows women from a wide range of ethnic and socio-economic
Mammographic image processing 7
In the monoenergetic simulation, the theoretical primary Table 2. Differences between linear attenuation coefficients over
energy [Equation (2)] becomes energies of interest.
image itself. This is important because knowledge of the in the computation of h int and for any reliable image analysis
breast thickness allows an estimate of the radiation dose during (see section 7.1.2).
the examination, and, in fact, the h int representation allows a The mammographic image can be corrected by using the
far more accurate estimation than is currently possible since data obtained by taking a blank image at low exposure with
it contains information specific to each breast, and is not a no ‘object’ present. As described earlier, this presents a dark
generic breast model. mammogram that includes the spatial variation of the incident
intensity. We correct for the anode heel effect in the energy
5. IMAGE RESTORATION AND ENHANCEMENT imparted representation, i.e. after converting film density to
energy imparted. The corrected energy imparted is
Once all the imaging conditions have been removed and the h int
representation computed, any projective X-ray examination φ(Vtube , xanode , yanode )
corrected
E ps (x, y) = E ps (x, y)
can be simulated. This enables us to produce images which φ(Vtube , x, y)
show the removal of known degrading factors such as scattered blank
E ps (xanode , yanode )
radiation and beam hardening. Because the algorithms are = blank (x, y)
E ps (x, y)
E ps
based on the physics of the imaging process they are less likely
to introduce artefacts than conventional algorithms. Since where E blank (x, y) are the energies imparted to the screen
users understand how the algorithms work they are better when the blank film is performed and (xanode , yanode ) is the
placed to diagnose from the enhanced image. Of course, position on the film directly below the anode, and is where the
as we noted earlier, the information contained intrinsically maximum film density on the blank film is.
in a mammographic image is limited (projective, high noise
to signal, poor discrimination of tissue type), but our work 5.2. Simulating a change of time of exposure
demonstrates that there is information on the films that is The time of exposure is set for each mammographic examina-
not usually seen so that we can be sure of performing what tion by an automatic exposure control (AEC) which terminates
radiologists would term image enhancement. However, note the exposure once a certain amount of radiation has been
that we do not seek to perform exact simulations: our goal absorbed. Occasionally, the AEC will misfunction producing
is to enhance the image, so we do not, for example, seek to images that are under- or overexposed, or the tissue above the
amplify the noise just in order to be accurate. AEC will be different between left and right breasts leading to
In this section we look at four algorithms, these simulate: an different exposures and making it hard to compare the images.
even incident radiation intensity; a change in time of exposure; In both cases, it is possible to simulate a change in exposure
removing the effects of scatter; changing the X-ray source. time, and hence a change in image brightness. Note the point
The first two of these can be simulated without going to made in the Introduction: the radiographer understands the
the h int representation but are both required for the last two effect of changing the exposure time, hence trusts using the
algorithms, which are the major innovations and are based corresponding Xmammo button and the algorithm it calls.
directly upon the h int representation. Further algorithms have There are two ways that the radiographer might change
been developed which allow the user to simulate any film– the time of exposure. One is through direct manipulation of
screen characteristic curve (Highnam, 1992) and the use of all the mA s (milli-Ampere seconds) value given at the time of
the algorithms can be seen in the accompanying CD-ROM, or exposure, the other is through manipulation of the AEC—
on the video available from the authors. both its position and density setting. Under- or overexposure
Note that although we discuss film density images, which of the image can be as a result of poor positioning of the AEC.
is where the pixel values are related linearly to film density, all Modelling the AEC as giving an average energy imparted
the images that we display are transmitted light images; that over a certain area has been shown to be a good model. To
is, the displayed images are adjusted so that the luminance simulate a change in the positioning of the AEC, one computes
from the monitor is directly proportional to the intensity of the current average energy imparted at the new area, E got
the light transmitted through the film when it is placed on a (AEC), and use the required energy E desired (AEC) (computed
light box. by specifying a desired average film density) to compute the
required increase in time of exposure:
5.1. Simulating an even incident radiation intensity
The ‘anode heel effect’ arises because of the way in which the tsdesired E desired (AEC)
X-rays are generated and leads to spatially varying incident got = .
ts E got (AEC)
radiation intensity over the breast. Although not necessarily a
degrading effect, it is necessary to be able to compensate for it All the energies in the image are multiplied by this ratio, and
Mammographic image processing 11
Figure 13. Distributions of mean values in small windows from film density and h feature images of three breasts. Around 100 samples were
taken for each example. The values computed on the transformed images show better clustering and differentiation of tissue types.
consider h min to be the surrounding level. The correct way of higher. This suggests using the average value within some
estimating h surround depends upon the definition of surrounding window on the image as a feature: higher averages tend
tissue that is being used and this, in turn, depends upon the type to indicate denser tissues. Figure 13 shows the distribution
of feature required, that is, we have to build in some notion of of average pixel value in small windows (1.5 mm × 1.5
scale. These issues are addressed in the next section. mm) in the film density and h feature images for the cyst and
fatty tissue in the three examples. Clearly, the average pixel
values in the film density images for the fatty tissues are not
8. FEATURE EXTRACTION sufficiently clustered to differentiate purely by thresholding.
For the h feature measures the value of h surround was taken to be
Due to the range of features and feature sizes in mammograms the minimum in the large window over which measurements
it is inappropriate to talk about general feature detection. For are performed. This is appropriate because it defines the
this paper we deal with features pertinent to detecting masses, surrounding tissue. In practice one might use a large window
in particular cysts, from fatty tissue, looking for feature values size that is greater than the maximum size of a cyst to
that are consistent within an image and between images. The ensure that some surrounding tissue is covered by the window.
three example images we use are meant to be illustrative of The fat values and the cyst values are far better clustered,
our approach; proof of superiority of our approach has to be in particular, the fatty tissue and the cysts in the breast at
statistical and that is beyond the scope of this paper. However, different compressions give almost identical responses, which
even with the relatively simple problem our approach appears we would expect given the compression model M3.
to have the advantage over a conventional approach. Fatty tissue, in particular, has low tissue roughness (there
The three examples that we consider are of cysts. Two are not many hills in the h int representation). There are many
of the examples (FDLC1,FDLC2) are of a cyst at different possible measures of roughness many of which are based
breast compressions. FDLC1 has the breast compressed to upon the concept of visual contrast so that the feature gives
5.55 cm with an exposure of 70.5 mA s, whilst FDLC2 has the high response for situations where we have a relatively high
breast compressed to 6.46 cm with an exposure of 92.7 mA s. peak compared with the background. Although such measures
The cyst in these images has surrounding tissue that appears might be appropriate for emulating a radiologist, we are more
mostly fatty. The third example (GCRC1) has a smaller cyst in interested in a measure of the tissue roughness that measures
an image digitized using a different scanner to the other two. the actual tissue property. For this paper we again use a simple
The breast was 5.80 cm thick with an exposure of 89.0 mA s. measure, this time standard deviation. The simplicity of our
Again, the surrounding tissue was relatively fatty. features comes in part from the task we are considering but
The first useful observation about cysts is that they often also from not needing to do any further normalization. Figure
present as being denser than fatty tissue. In this case, the 14 shows the standard deviation distribution from the three
attenuation is higher and consequently the image brightness is
Mammographic image processing 17
Figure 14. Distributions of standard deviation values in small windows from film density and h feature images of three breasts. Around 100
samples were taken for each example. Note that the horizontal axis are marked differently, this reflects the fact that the original images are
integers whilst the transformed images are floats.
film density and h feature images for the cysts and surrounding • Image matching. The h int representation suggests new
fatty tissue. The distributions show that the average of the approaches to difficult problems such as matching two
standard deviations is lower for the fatty tissue and that the such structures: images of the same breast taken at
distributions do overlap for the film density images. However, different times, two different views of the same breast,
the distributions in the h feature images are closer than those in comparison of a woman’s left and right breasts, or
the film density images, especially in the case of the breasts at comparison of lots of breasts from the same view. The
different compressions. It would appear that the h feature values development of image matching techniques based on
are more robust to breast compression. the h int representation, and the comparison of results to
image based techniques (Bowyer et al., 1992; Sallam and
9. FUTURE WORK Bowyer, 1994) will be one of our earliest developments
of the research reported here.
In this paper we have introduced the h int surface representation, • Computing mass density. It might be possible to calculate
shown how to compute it, and illustrated its use in a number automatically the proportion of interesting tissue-to-fat
of applications in mammographic image processing. Current for a mass by matching predicted appearance to actual
work aims to extend that described here in many ways: appearance. The spatial size of the tumour is known from
its projected shape and using a suitable tumour model a
• Surface feature extraction. We noted earlier that the
three-dimensional shape can be proposed. The computer
h int representation is a surface representation of breast
then changes the proportion of interesting tissue-to-fat
anatomy, freed of the imaging choices that are con-
until the pixel values in the predicted and actual images
founded in regarding the image as a surface. On the
match.
h int surface, cancerous masses correspond to hills, for
which the steepness of the hill slope, its height relative to • Computing radiation dose. The h int representation
the surrounding ‘plain’ and the jaggedness of its outline supports an estimate of the radiation dose during the
are important parameters for distinguishing malignant examination that is more accurate than is currently
from benign cancers. This line of thinking leads to the possible. Current techniques use a standard breast model,
evaluation of techniques to extract useful information we can use details from the specific breast. Furthermore,
robustly from digital surfaces, such as first, second and we can consider images over time and match them to
even higher order surface variations, including lines of compute the combined radiation dose to certain tissues.
curvature and crest lines, all at multiple scales. We aim • Integrating MRI and X-ray. Using h int we can integrate
to apply such techniques to the h int surface to extract and MRI and X-ray of the breast to investigate hardness and
evaluate potential masses. mobility of tissue, and to investigate recidivist masses.
18 R. Highnam et al.
From an MRI image of the breast it is possible to classify Dance, D. R. and Day, G. J. (1984) Computation of scatter in
each voxel into one of our tissue classifications, and an mammography by Monte Carlo methods. Phys. Med. Biol., 29,
X-ray simulated. The differences between the simulated 237–247.
and actual X-ray images are due to the compression of Dance, D. R., Persliden, J. and Carlsson G. A. (1992) Calculation of
the breast in the X-ray image, and these differences can dose and contrast for two mammographic grids. Phys. Med. Biol.
37, 235–248.
thus give us an idea about the hardness and mobility
Day, G. J. and Dance, D. R. (1983) X-ray transmission formula for
of the tissue. If tissues can be matched between the antiscatter grids. Phys. Med. Biol., 28, 1429–1433.
two images, it might be possible to more accurately Forrest, P. (1986) Breast Cancer Screening. Report to the Health
differentiate between benign and malignant tissues Ministers of England, Wales, Scotland and Northern Ireland.
(Highnam et al., 1991). Finally, if a cancer is detected, HMSO.
leading to a partial mastectomy or lumpectomy, there Gilles, S. (1995) Analyse et Traitement d’Images Médicales Tri-
is an increased risk of subsequent recidivism, but dimensionelles: Application à la Détection Précoce du Cancer
such subsequent growths tend to be near the surgical du Sein. Master’s Thesis, Université Paris VII-Denis Diderot.
scar tissue and they are hard to detect using X-ray Haus, A. G., Metz, C. E., Doi, K. and Bernstein, J. (1977)
mammography alone. We have recently developed Determination of x-ray spectra incident on and transmitted
a prototype system (Gilles, 1995) to analyse contrast through breast tissue. Radiology, 124, 511–513.
Highnam, R. P. (1992) Model-Based Enhancement of Mammo-
enhanced MRI images for application to early detection
graphic Images. Ph.D. Thesis, Department of Engineering Sci-
of breast cancer in young, at risk groups of women. ence, Oxford University.
Highnam, R. P., Shepstone, B. J. and Brady, J. M. (1991) Mammo-
ACKNOWLEDGEMENTS grams at Different Compression Plate Widths for Detection of
Breast Cancer. In Radiology and Oncology 91, Work in Progress
The authors thanks the staff at the Churchill Hospital, Oxford,
(abstracts), 3. British Institute of Radiology.
for their continuing support and encouragement. Particular
Highnam, R. P., Brady, J. M. and Shepstone, B. J. (1994) Computing
thanks to Yvonne Swainston, Anne Dickson-Browne, Donald the scatter component of mammographic images. IEEE Trans.
Peach, and Niall Moore. Nick Cerneaz did the early work Med. Imaging, 13, 301–313.
on the curvilinear structures. Thanks to Alison Noble and Highnam, R. P., Brady, M. and Shepstone, B. (1995) A Represen-
Gabor Szekely for comments on an earlier draft. R.P.H. tation for Mammographic Image Processing. In First Int. Conf.
thanks Philips Laboratories for support during his 9 month on Computer Vision, Virtual Reality and Robotics in Medicine,
stay there and Chuck Carman for many useful discussions; CVRMed’95, Lecture Notes in Computer Science, Ayache, N.
J.M.B. thanks the Epidaure Team at INRIA Sophia Antipolis, (ed.), 905. Springer-Verlag, Nice.
and Nicholas Ayache in particular, for tremendous support Johns, P. C. and Yaffe, M. J. (1987) X-ray characterisation of normal
during his sabbatical there. and neoplastic breast tissue. Phys. Med. Biol., 32, 675–695.
Karssemeijer, N. (1993) Adaptive noise equalization and recognition
of microcalcification clusters in mammograms. Int. J. Patt. Recog.
REFERENCES
Artif. Intell., 7, 1357–1376.
Sallam, M. and Bowyer, K. (1994) Registering Time Sequences
Bowyer, K., Sallam, M., Hubiak, G. and Clarke, L. (1992) Screening
of Mammograms Using a Two-Dimensional Image Unwraping
Mammogram Images for Abnormalities Developing Over Time.
Technique. In 2nd International Workshop on Digital Mammog-
In IEEE Nuclear Science Symp. and Medical Imaging Conf.,
raphy, Excerpta Medical International Congress Series 1069 ,
1270–1272.
Gale, A. G., Astley, S. M., Dance, D. R. and Cairns, A. Y. (eds),
Burch, A. and Law, J. (1995) A method for estimating compressed
121–130. Elsvier Science, York.
breast thickness during mammography. Brit. J. Radiol., 68, 394–
Shroff, J. H., Lloyd, L. R. and Schroder, D. M. (1991) Open breast
399.
biopsy: a critical analysis. Am. Surgeon, 57, 481–485.
Carlsson, G. A., Dance, D. R. and Persliden, J. (1989) Grids
Tarassenko, L., Hayton, P., Cerneaz, N. and Brady, M. (1995) Novelty
in Mammography: Optimization of the Information Content
Detection for the Identification of Masses in Mammograms. In
Relative to Radiation Risk. Technical Report ULi-RAD-R-059,
Fourth Int. Conf. Artif. Neural Networks, 442–447. Institute of
Linkoping University, Department of Radiation Physics.
Electronic Engineers, Cambridge.
Cerneaz, N. J. (1994) Model-Based Analysis of Mammograms. Ph.D.
Woodman, C. B., Threlfall, A. G., Boggis, C. R. and Prior, P. (1995) Is
Thesis, Department of Engineering Science, Oxford University.
the three year breast screening interval too long? Occurrence of
Cerneaz, N. J. and Brady, J. M. (1995) Finding Curvilinear Structures
interval cancers in the NHS breast screening programme’s north
in Mammograms. In First International Conf. on Computer
western region, Brit. Med. J., 310, 224–226.
Vision, Virtual Reality and Robotics in Medicine, CVRMed’95,
Lecture Notes in Computer Science, Ayache, N. (ed.), 905.
Springer-Verlag, Nice.