Iranian Journal of Medical Physics

Vol. 11, No. 2 & 3, Spring & Summer 2014, 260-269

Received: April 9, 2013; Accepted: October 1, 2013

Original Article

Contrast Enhancement of Mammograms for Rapid Detection of

Microcalcification Clusters

Hajar Moradmand 1, Saeed Setayeshi 1, Alireza Karimian2*, Mehri Sirous3

Abstract

Introduction
Breast cancer is one of the most common types of cancer among women. Early detection of breast cancer is
the key to reducing the associated mortality rate. The presence of microcalcifications clusters (MCCs) is one
of the earliest signs of breast cancer. Due to poor imaging contrast of mammograms and noise
contamination, radiologists may overlook some diagnostic signs, specially the presence of MCCs. In order to
improve cancer detection, image enhancement methods are often used to aid radiologists. In this paper, a
new enhancement method was presented for the accurate and early detection of MCCs in mammograms.
Materials and Methods
The proposed system consisted of four main steps including: 1) image scaling;2) breast region
segmentation;3) noise cancellation using a filter, which is sensitive to MCCs; and 4) contrast enhancement of
mammograms using Contrast-Limited Adaptive Histogram Equalization (CLAHE) and wavelet transform.
To evaluate this method, 120 clinical mammograms were used.
Results
To evaluate the performance of the image enhancement algorithm, contrast improvement index (CII) was
used. The proposed enhancement method in this research achieved the highest CII in comparison with other
methods applied in this study. The Validity of the results was confirmed by an expert radiologist through
visual inspection.
Conclusion
Detection of MCCs significantly improved in contrast-enhanced mammograms. The proposed method could
be helpful for radiologists to easily detect MCCs; it could also decrease the number of biopsies and reduce
the frequency of clinical misdiagnosis. Moreover, it could be useful prior to segmentation or classification
stages.
Keywords: Breast Cancer, Image Enhancement, Wavelet Transform.

1- Department of Biomedical-Radiation Engineering, Amirkabir University of Technology, Tehran, Iran

2- Department of Biomedical Engineering, Faculty of Engineering, University of Isfahan, Isfahan, Iran
*Corresponding author: Tel: +98 313793-4059; E-mail: [email protected]
3- Department of Radiology, Isfahan University of Medical Sciences, Isfahan, Iran

Iran J Med Phys, Vol. 11, No. 2 & 3, Spring & Summer 2014 260
Enhancement of Mammograms for Rapid Detection of MCCS

1. Introduction on the skills of a radiologist, the reader’s

Breast cancer is one of the major causes of sensitivity can be increased by approximately
mortality among women. In Iran, patients with 10-15% with the assistance of CAD systems
breast cancer are at least one decade younger [5].
than their counterparts in developed countries; A CAD system generally consists of four
i.e., age of breast cancer development has stages of preprocessing, image segmentation,
reduced from 40 to 30years [1].This is a feature extraction, and classification of
serious problem in our country and further features [5].Preprocessing in the CAD system
research is required in this area. Since causes significantly affects other stages. Effective
of breast cancer remained unknown, there is preprocessing decreases the error rate in the
no an assured way to prevent it; therefore, the next steps, and thus, the total number of errors
best way to fight this disease is early diagnosis in the system is reduced. With proper
in early stages of the disease. preprocessing, the minimum number can be
The appearance of microcalcifications (MCs) selected as primary suspected MCCs; this
in mammograms is one of the first signs in 30- would accelerate the classification stage.
50% of breast cancers [2]. MCs are tiny specks MCs appear in very small sizes and various
of calcium in the breast, and are produced due patterns on mammography images. Therefore,
to various reasons including breast cancer. it is very difficult to identify them if they are
MCs can be often seen in malignant tissues as not formed as concentrated clusters. Also,
cluster patterns. Microcalcification clusters factors such as the large size and the low
(MCCs) in 60-80% of breast cancers have contrast of mammograms, as well as the small
been observed after histological examination size, various patterns, and limited number of
[3]. Considering the significant correlation MCCs, can increase the processing time in
between the presence of MCCs and breast CAD systems. To overcome the mentioned
cancer, early detection of these signs on problems, image noise must be removed and
mammograms can help control breast cancer. the contrast between the background region
Generally, MCs have small sizes (range of and region of interest (ROI) must be improved.
about 0.1mm to 0.7 mm) and low contrast; All these operations are performed in the
therefore, they may be overlooked or preprocessing stage.
misinterpreted by radiologists. In fact, eye- Many algorithms have been proposed to
based mammography screening is a tedious improve the quality of mammograms
task, and about 10-30% of this signs is including histogram modification, median
overlooked by physicians [4]. Physicians are filters, Gaussian filters, morphological filters
responsible for the final confirmation of a [6], and fractal models [7]. Recently,
disease, thus, due to misdiagnosis, many researchers have paid particular attention to
patients may undergo unnecessary biopsies, multi-scale techniques in order to improve the
which result in the patient’s pain and image contrast, and wavelet-based methods are
unnecessary costs. Moreover, lack of early among the highly efficient methods [8].
diagnosis can result in the patient’s death, and Although some of these approaches improve
under the best circumstances, it may lead to the visibility of MCCs, the layout and shape of
the complete removal of breast tissue. these calcium deposits may be distorted in
Considering the difficulties and the importance images, which may lead to misdiagnosis.
of detecting MCCs in breast cancer diagnosis, Some other methods may increase the noise or
the need to use automatic methods is strongly destruction to anatomical structures of image
felt in order to prevent the development of in the image. The size and shape of diagnostic
cancer cells. Recently, computer-aided indicators significantly change in
diagnosis (CAD) systems have improved the mammograms and conventional methods are
detection of abnormalities in mammograms, weak in improving various features of
and are used as a “second opinion”. Depending mammograms. Therefore, adaptive contrast

261 Iran J Med Phys, Vol. 11, No. 2 & 3, Spring & Summer 2014
 Hajar Moradmand et al.

enhancement algorithms, in which the focus is method called image scaling [14], the size of
on the local content of image, may be helpful images was reduced without losing any
[9]. significant features of mammograms [15, 16];
Each of these methods has its own advantages image scaling was performed as the first step
and disadvantages for improving mammogram in the process.
quality. The aim of this study was to present a 2.2 Breast region extraction
new method for improving image contrast in Breast tissue occupies only a part of the image
order to detect MCCs earlier and more and the rest of the image is all black,
accurately. This method can also be used in containing no information. In a mammogram,
the later stages of CAD system such as the types of noise include high-intensity labels,
segmentation and classification of benign and low-intensity labels, and tape artifacts. In order
malignant patterns [10]. to increase the processing speed and reduce
The proposed method in this study included the effects of artifacts, we tried to extract the
four stages. Firstly, the images were scaled breast region with a mask template.
since the real size of mammograms was too Consequently, subsequent processing was
large. Secondly, the breast tissue was performed only on breast tissue.
segmented from the image background. The mask template was a binary matrix with a
Thirdly, noise was removed using a sensitive size equal to that of the original image.
filter to the shape and size of MCCs (an Viscosity and density of breast tissue and fat
adaptive median filter was used for this were not similar among different individuals.
purpose). Finally, the combined method of Therefore, applying the conventional threshold
Contrast-Limited Adaptive Histogram could not provide the desired mask template to
Equalization (CLAHE) and wavelet transform properly separate all breast tissue. To
was used for contrast enhancement. We overcome this problem, first, a series of
selected the algorithms based on previous morphological operations including erosion
studies and tried to eliminate the associated and dilation were performed to smooth the
disadvantages to identify features such as the mammogram; then, a global threshold was
presence of MCCs [11-13]. defined.
Intensities less than the defined threshold were
2. Materials and Methods set to 0 (black) and higher intensities were set
2.1 Data bank to 1 (white); thus, the images were converted
All mammograms in this study were obtained to binary images. However, the visible noises
from patients, referring to a radiology center in on the mammograms, as mentioned above, still
Isfahan, Iran, in years 2009 - 2011. The data remained outside the zone. To select only the
included 120 mammograms containing one or breast region for further processing,
more MCCs. The mean size of images was morphological operations were used to remove
4000×6000, with a sample rate of 50 mm and areas smaller than 1000 pixels.
resolution of 8 bits per pixel. These images Among the remaining regions, the largest area
were converted to digital images using a laser was selected as a border area between the
digitizing system. breast and the black part of the image. An
The large size of these images made the example of an applied mask on the image is
processing, storing, and sending processes shown in Figure 1.
very time-consuming and difficult. By using a

Iran J Med Phys, Vol. 11, No. 2 & 3, Spring & Summer 2014 262
Enhancement of Mammograms for Rapid Detection of MCCS

Figure 1. An example of the applied mask on the extracted breast tissue

2.3 Noise cancellation 2.4 Adaptive median filter

Digital mammograms often contain a An adaptive median filter implements a spatial
substantial amount of impulse noise, which is operation to determine which pixels of the
observed as bright or dark spots (salt or pepper image are affected by impulse noise. By
noise) on the image. comparing each pixel in the image with the
So far, various filters have been proposed to neighborhood, the pixel is classified as noise;
reduce this type of noise. The first category of the window size of the neighborhood changes
these filters has a good performance in during the operation. Suppose that the image
reducing image noise, but it typically destroys noise is denoted by y and S , is a window with
image details and makes the image blurry. The w × w size, located in the center of(i, j),
second category of these filters preserves which:
image details, though noise reduction is low. S,
The third category removes noise while = {(k; l): |k − i| (1)
preserving image edges and details. ≤ w and|k − j| ≤ w }
Median filter is included in the third category Minimum, maximum, and median gray levels
of these filters. This non-linear filter is a in the “w” window are calculated, which have
powerful and useful tool for improving the the following notations:S , , : minimum
quality of mammograms, given its ability to
gray level in the w, S , , : maximum gray
remove impulse noise without blurring the
edges [17].The median filter uses a m×n level in the w, S , , : median gray level in
neighborhood and arranges the whole the w, Y , : gray level at the (i, j) point of
neighborhood in an ascending order. Then, it image, and w : the maximum window size
selects the median of the sorted numbers and of the median filter.
substitutes it for the central pixel. The window size is set in a way that S , , is
If the image details are not sufficient or the
gray levels of pixels are very different from not an impulse noise. If Y , is the impulse
each other, the image is considered blurred noise, it is replaced with S , , . , otherwise,
despite the use of the median filter. In other the output remains unchanged (Y , ) and is not
words, a simple median filter cannot make a considered an impulse noise [18]. Block
distinction between noise and fine details of diagram of the proposed algorithm for the
the image, and it should be removed. adaptive median filter is shown in Figure 2.
In this research, adaptive median filter, as a
more suitable filter compared to a simple
median filter, was used for noise cancellation.

263 Iran J Med Phys, Vol. 11, No. 2 & 3, Spring & Summer 2014
 Hajar Moradmand et al.

described. For more information, readers can

refer to the reference section [19-21].
2.6.(CLAHE)
Mammogram histogram is considered as a
distribution probability, based on the
information theory. A mammogram with a
uniform histogram can provide more
information. Maximum entropy can be
achieved by histogram equalization, although
normal tissues and noise will be amplified.
CLAHE can be utilized in a small area of the
image (called tiles) rather than the entire
image. The contrast of each of these areas is
enhanced so that the output corresponds to a
specific histogram. Then, the histogram tiles
are matched together and form the final image;
this could prevent any probable noise
amplification in the image [19].
CLAHE is an improved version of Adaptive
Histogram Equalization (AHE).CLAHE
algorithm partitions the images into contextual
regions and applies the histogram equalization
for each region; this uniforms the distribution
of gray values and thus makes hidden features
Figure 2. Block diagram of the proposed algorithm of
of the image more visible. The contrast,
adaptive median filter especially in homogeneous areas, can be
limited in order to avoid amplifying any noise
2.5 Contrast enhancement that might be present in the image. An
In this study, a combination of methods example of the performance of histogram
(CLAHE and the wavelet transform) was used equalization is shown in Figure 3.
to enhance image quality. In the following
section, the performances of these methods are

a b c
Figure 3.An example of the performance of CLAHE on a mammogram, a) original mammogram, b) enhanced
mammogram after applying histogram equalizations, c) enhanced mammogram after applying CLAHE

Iran J Med Phys, Vol. 11, No. 2 & 3, Spring & Summer 2014 264
Enhancement of Mammograms for Rapid Detection of MCCS

2.7 Wavelet transform Coiflet (coifN) wavelets are a family of

A wavelet transforms a signal into a set of orthogonal wavelets, which are suitable for
basic functions, which vary in terms of scale mammograms. Based on the fact that wavelet
factor (s) and position (b), and are derived coefficients represent the degree of similarity
from the mother wavelet ψ (t). Wavelet between the image and the mother wavelet, the
transform of the signal f (t) is defined as following equations are used to assess the
follows: similarity between coefficients:
1 t−b ∑
Eh = 100 ∑ , Ev = 100 ∑ ,Ed =
∑
Wsf(b) = f(t)ψ dts
√s s (2) ∑
>0 100 ∑ (3)
The two-dimensional wavelet transform is
applied in image processing. Two-dimensional In this study, five-level discrete wavelet
wavelet transform is obtained by decomposition was employed using
multiplication of the scale (ψ) and the wavelet Daubechies wavelet of order 8,since it
function (φ). A two-dimensional wavelet accumulates more energy corresponding to the
decomposes an image into one image with details of wavelet transform [21];moreover, it
approximation coefficient and detail is characterized by symmetry and finite length
coefficients in three orientations (horizontal, to enhance mammograms. Considering these
vertical, and diagonal). An approximate features, this decomposition can attain a high
coefficient contains large-scale components, correlation with the clustered MCs, and
though detail coefficients contain small-scaled therefore, it can enhance the visibility of
ones; in frequency-domain analysis, detail MCCs.This produced an approximate
coefficients are high-frequency coefficients. coefficient and five sets of horizontal, vertical,
MCCs frequently appear on mammograms as and diagonal detail coefficients. After
fine bright grains in the breast tissue. It can be decomposition, the algorithm performed
assumed that with wavelet decomposition, inverse wavelet transforms using these five
MCCs are often among detail coefficients in sets of detail coefficients, and the approximate
mammograms [20]. This assumption is the coefficients were set to zero.
basis for the proposed algorithm to improve
the image contrast.
3. Results
Several families of wavelets can be selected
The proposed method in this study was applied
for the processing of mammography images.
for120 digitized mammograms. Two examples
For these images, the orthogonality is the main
of applying the proposed algorithm for
criterion for selecting a family. Orthogonal
contrast enhancement of images, containing
properties allow the input image to be
MCCs, are presented in Figure 4.
decomposed into independent coefficients.
The Daubechies (dbN), Symlet (symN), and

265 Iran J Med Phys, Vol. 11, No. 2 & 3, Spring & Summer 2014
 Hajar Moradmand et al.

Figure 4. Two examples of the proposed algorithm for two different mammograms, a) original ROI of the mammogram,
(b) enhanced ROI
The contrast C of a region was defined by the
As shown in Figure 4, the proposed algorithm following equation:
could improve image contrast, and thus, the f−b (5)
MCCs were visible. c=
f+b
To evaluate the performance of the proposed Where f is the mean gray-level value of the
algorithm, the contrast improvement index foreground, and b is the mean gray-level value
(CII) was used [22], which is defined as of the background.The higher value of CII
follows: shows the best performance of the
C (4) enhancement method. In Table 1, the contrasts
CII =
C of the original and enhanced images are
Where C and C represent the indicated.
contrast values of the processed and original
images, respectively.
Table 1.The CII values of four different enhanced mammograms
Contrast Original Wavelet CLAHE The proposed
transform algorithm
C 0.1331 0.2054 0.2106 0.3596
Image1 CII - 1.54 1.58 2.7
C 0.1385 0.2367 0.2660 0.4533
Image2
CII - 1.70 1.92 3.27
C 0.1193 0.1850 0.2063 0.4279
Image3
CII - 1.55 1.73 3.58
C 0.1151 0.2738 0.3021 0.4216
Image4
CII - 2.37 2.62 3.66

Iran J Med Phys, Vol. 11, No. 2 & 3, Spring & Summer 2014 266
Enhancement of Mammograms for Rapid Detection of MCCS

To assess the quality of an image after to stones, whose sizes can be determined by
enhancement, profile intensity surface area sifting them through screens of increasing size
distribution curve was used. The intensity and collecting what remains after each pass.
distribution curve was applied for the Image objects are sifted by opening the image
measured intensity distribution of MCCs as a with a structuring element of increasing size
function of size without any need for and counting the remaining intensity surface
separating MCCs. This curve estimates the area (summation of pixel values in the image)
intensity surface area distribution of MCCs as after each opening [23]. The intensity curve of
a function of size and compares image objects a mammogram is shown in Figure 5.

Figure 5.Intensity surface area distribution curve

for the physicians. Considering the difficulties
As shown in Figure 5, the proposed algorithm and the importance of detecting MCCs in
achieved the highest intensity compared to the breast cancer diagnosis, the need to use
CLAHE and wavelet transform. This asserts automatic methods is strongly felt in order to
that the proposed method in this research can prevent the development of cancer cells.
effectively improve the image contrast. In this study, a new method for improving
mammography image contrast was presented
4. Discussion in order to detect MCCs earlier and more
The presences of MCCs are one of the earliest accurately. The proposed method included
and most common signs of breast cancer. four stages. Firstly, the images were scaled
Detection of these signs in mammograms is since the real size of mammogram images was
more important than any other symptoms. too large. Secondly, the breast tissue was
Factors such as the large size and the low segmented from the background of the image.
contrast of mammograms, as well as the small Thirdly, noise was removed using a sensitive
size, various patterns, and limited number of filter in correspondence with the shape and
MCCs, make the detection of MCCs difficult size of MCCs (an adaptive median filter was

267 Iran J Med Phys, Vol. 11, No. 2 & 3, Spring & Summer 2014
 Hajar Moradmand et al.

used for this purpose). Finally, the combined effect on contrast enhancement of
method of CLAHE and wavelet transform was mammograms, compared to manual methods.
used for contrast enhancement. The proposed method showed a more
We selected the algorithms based on previous significant effect on dense breast tissue;
studies and tried to eliminate the associated however, the original images in some fatty
disadvantages to identify features such as breast tissue were preferred.
MCCs[11-13].To evaluate the performance of In general, the proposed algorithm could
the proposed algorithm, CII was used. The improve the physicians’ diagnosis based on
mean CII for the proposed algorithm in this mammograms, decrease the number of
study was 3.34 and for the wavelet and biopsies, and reduce the frequency of clinical
CLAHE was 1.81 and 1.98, respectively. This misdiagnosis.
indicates that the suggested algorithm Although the study results were satisfactory,
achieved the highest CII among others. the effective use of this model in medical
Two radiologists also evaluated the images, centers needs further studies on more images.
obtained by the proposed method. The results In future works, these results could be used in
showed that this method could improve the subsequent stages of CAD system such as
accuracy and speed of MCCs detection. To segmentation and classification of benign and
implement the presented method, MATLAB malignant patterns.
software version 7.4 was used, which is
applicable for conventional systems. Acknowledgment
The authors would like to thank all the
5. Conclusion participants who cooperated with us in this
According to the results, the most useful study. We also express our deepest gratitude to
characteristic of this system was the automatic Dr. Mohammad Esmail Akbari,the chief of
contrast enhancement; however, although this Iranian Cancer Research Center, for his
method manually improved the image contrast, sincere contributions and Mr. Hossein
it was time-consuming. Khazaei, the reviewer of the article, for his
In addition, it was believed that the proposed valuable comments.
method in this paper has a more significant

References
1. Zare N, Haem E, Lankarani KB, Heydari ST, Barooti E. Breast cancer risk factors in a defined population:
weighted logistic regression approach for rare events. J Breast Cancer. 2013 Jun;16(2):214-9.
2. Tang J, Rangayyan RM, Xu J, El Naqa I, Yang Y. Computer-aided detection and diagnosis of breast cancer
with mammography: recent advances. IEEE Trans InfTechnol Biomed. 2009 Mar;13(2):236-51.
3. Cheng H-D, Cai X, Chen X, Hu L, Lou X. Computer-aided detection and classification of microcalcifications
in mammograms: a survey. Pattern Recognition. 2003;36(12):2967-91.
4. Dean JC, Ilvento CC. Improved cancer detection using computer-aided detection with diagnostic and screening
mammography: prospective study of 104 cancers. AJR Am J Roentgenol. 2006 Jul;187(1):20-8.
5. Thangavel K, Karnan M, Sivakumar R, Mohideen AK. Automatic detection of microcalcification in
mammograms-a review.International Journal on Graphics Vision and Image Processing. 2005;5(5):31-61.
6. Nakayama R, Uchiyama Y, Yamamoto K, Watanabe R, Namba K. Computer-aided diagnosis scheme using a
filter bank for detection of microcalcification clusters in mammograms. IEEE Trans Biomed Eng. 2006
Feb;53(2):273-83.
7. Bocchi L, Coppini G, Nori J, Valli G. Detection of single and clustered microcalcifications in mammograms
using fractals models and neural networks. Med Eng Phys. 2004 May;26(4):303-12.
8. Ramírez-Cobo P, Vidakovic B. A 2D wavelet-based multiscale approach with applications to the analysis of
digital mammograms.Computational Statistics & Data Analysis. 2013;58:71-81.
9. Mohan S, Ravishankar M. Modified contrast limited adaptive histogram equalization based on local contrast
enhancement for mammogram images. Mobile Communication and Power Engineering: Springer; 2013. p.
397-403.

Iran J Med Phys, Vol. 11, No. 2 & 3, Spring & Summer 2014 268
Enhancement of Mammograms for Rapid Detection of MCCS

10. Papadopoulos A, Fotiadis DI, Costaridou L. Improvement of microcalcification cluster detection in

mammography utilizing image enhancement techniques. ComputBiol Med. 2008 Oct;38(10):1045-55.
11. Zhang X, Homma N, Goto S, Kawasumi Y, Ishibashi T, Abe M, et al. A Hybrid Image Filtering Method for
Computer-Aided Detection of Microcalcification Clusters in Mammograms.Journal of Medical Engineering.
2013;2013.
12. Moradmand H, Setayeshi S, Karimian AR, Sirous M, Akbari ME. Comparing the Performance of Image
Enhancement Methods to Detect Microcalcification Clusters in Digital Mammography.Iranian Journal of
Cancer Prevention. 2012;5(2):61-8.
13. Moradmand H, Setayeshi S, Khazaei H. Comparing Methods for segmentation of Microcalcification Clusters
in Digitized Mammograms.International Journal of Computer Science Issues. 2011; 8: 104-8.
14. González RC, Woods RE. Digital image processing: Prentice Hall; 2002.
15. Hashimoto B. Practical Digital Mammography: Thieme;2011.
16. Yu S, Guan L. A CAD system for the automatic detection of clustered microcalcifications in digitized
mammogram films.IEEE Trans Med Imaging. 2000 Feb;19(2):115-26.
17. Langarizadeh M, Mahmud R, Ramli AR, Napis S, Beikzadeh MR, Rahman WE. Improvement of digital
mammogram images using histogram equalization, histogram stretching and median filter. J Med Eng
Technol. 2011 Feb;35(2):103-8.
18. Yuan S-Q, Tan Y-H. Impulse noise removal by a global–local noise detector and adaptive median filter.Signal
Processing. 2006;86(8):2123-8.
19. Zeng M, Li Y, Meng Q, Yang T, Liu J. Improving histogram-based image contrast enhancement using gray-
level information histogram with application to X-ray images. Optik-International Journal for Light and
Electron Optics. 2012;123(6):511-20.
20. Mencattini A, Salmeri M, Lojacono R, Frigerio M, Caselli F. Mammographic images enhancement and
denoising for breast cancer detection using dyadic wavelet processing. Instrumentation and Measurement,
IEEE Transactions on. 2008;57(7):1422-30.
21. Matsuyama E, Tsai DY, Lee Y, Tsurumaki M, Takahashi N, Watanabe H, et al. A modified undecimated
discrete wavelet transform based approach to mammographic image denoising. J Digit Imaging. 2013
Aug;26(4):748-58.
22. Morton MJ, Whaley DH, Brandt KR, Amrami KK. Screening mammograms: interpretation with computer-
aided detection--prospective evaluation. Radiology. 2006 May;239(2):375-83.
23. Lira C, Pina P. Granulometry on classified images of sand grains. Journal of Coastal Research, ICS 2011
Proceedings, SI64. 2011..

269 Iran J Med Phys, Vol. 11, No. 2 & 3, Spring & Summer 2014