See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/340695850

A Computer Aided Detection System for Breast Cancer in the

MammogramsBased on Particle Swarm Optimization Algorithm

Conference Paper · December 2019

DOI: 10.1109/ICCES48960.2019.9068111

CITATIONS READS

0 73

4 authors:

Nesma Elsokkary Amany A. Arafa

Egyptian Atomic Energy Authority Egyptian Atomic Energy Authority
4 PUBLICATIONS   8 CITATIONS    21 PUBLICATIONS   57 CITATIONS   

SEE PROFILE SEE PROFILE

Ahmed Hamza Asad Hesham A. Hefny

Faculty of Graduate Studies for Statistical Research, Cairo University Faculty of Graduate Studies for Statistical Research (FGSSR)
20 PUBLICATIONS   132 CITATIONS    273 PUBLICATIONS   1,311 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Conference Paper Discovering the classification rules for Egyptian stock market using genetic programming View project

image watermarking and compression View project

All content following this page was uploaded by Ahmed Hamza Asad on 21 April 2020.

The user has requested enhancement of the downloaded file.

 A Computer Aided Detection System for Breast
Cancer in the MammogramsBased on Particle
Swarm Optimization Algorithm
Nesma El-Sokkary A. A. Arafa Ahmed H. Asad Hesham A. Hefny
Department of Radiation Department of Radiation Computer Science Computer Science
Engineering, National Center Engineering, National Center Department, Faculaty of Department,Faculaty of
for Radiation Research and for Radiation Research and Graduates Studies for Graduates Studies for
Technology (NCRRT) Technology (NCRRT) Statistical Research Statistical Research
Egyptian Atomic Energy Egyptian Atomic Energy Cairo University Cairo University
Authority Authority Cairo, Egypt Cairo, Egypt
Cairo, Egypt Cairo, Egypt [email protected] [email protected]
nesma.elsokkary@ gmail.com [email protected]

Abstract—the majority cancer mortality among women is highly experienced. While, other countries examination
due to breast cancer over the world wide. Recent researches system approves the single reading in the screening method
have shown the effectiveness of x-ray mammography in early [5]. The early detection of cancer is a complex task that
detection of breast cancer. Unfortunately, the present systems requires scanning and screening huge number of images.
for early detection are expensive and needs extremely complex Hence, it is easy to make mistakes and miss classifying
algorithms. The crucial challenge in designing a computer- cancer cases.
aided detection (CAD) systems for breast cancer are the
segmentation phase, which requires highly complex The accuracy of detecting and diagnosing mammography
computation. Hence, this paper proposes a CAD system to be rely on many factors like breast density, breast structure and
utilized for breast cancer detection in mammographic datasets. other factor related to the radiologist himself such as his
The segmentation step is performed by a Particle Swarm opinion, skills and experience. It is predetermined that 70%
Optimization Algorithm (PSO). Statistical, textural and shape of all missed breast cancer are due to misunderstanding,
feature are calculated over the segmented region. A non linear while the other 30% are due to disregard lesions [6].
support vector machine (SVM) is exploited in the next phase in Moreover, there is a deficiency in the trained radiologists
order to analyze the extracted features and classify the cause a growing attention in analyzing the mammogram
mammograms into normal, benign or malignant. For the sack images using computerized way.
of evaluating the performance, the experiment is performed on
Mini-MIAS database . The obtained accuracy rates based on Therefore, the need for developing the Computer-aided
10-folds cross validation are 85.4% for classifying normal from detection (CAD) systems become indispensable in order to
abnormal, 89.5% for classifying malignant from benign. The enhance the sensitivity of mammography screening. The
experiment shows that the classification accuracy is 81% when sensitivity is mainly increased by specifying suspected
classifying normal, malignant or benign. The result regions on mammograms such as micro calcifications and
compromises with recent researches concurs that the proposed tumors [7, 8]. When applying the CAD system in cancer
algorithm compromises between the achieved accuracy to detection, only one radiologist will be responsible for
complexity cost. approving the concluded decision and has to decide to recall
a woman or not to perform a further physical examinations.
Keywords—breast cancer, CAD system, Mammography, PSO,
Segmentation, SVM.
[9]. The CAD-based assessment possibly depends on a single
referred area that the radiologist didn't notice at first. In
I. INTRODUCTION addition, CAD system has been recommended to be used for
training and educated the radiologists [10].
The biggest threat that faces the women in our
community is the breast cancer. Regarding the results of This paper studies and proposes a method for computer-
National Cancer Registry Program (NCRP) [1] in Egypt, the aided detection of breast abnormalities that would assist the
breast cancer event rate per 100,000 cases were calculated early discovery of the breast cancer in mammograms. The
and reported as 32.0% [2]. For all kinds of cancers, the prime proposed system aims to help radiologists by providing him
factor for enhancing the opportunity for recovery is the early with a guide decision for the patient’s status. The particle
detection of neoplasm. Computerized mammography swarm optimization algorithm is borrowed into the field of
screening program can assist the early diagnosing of breast segmenting the mammogram images for achieving a positive
cancer [3]. In mammography one is not looking at an identification of a breast abnormalities. In other words, the
individual cell level but at image level, which hopefully main contribution of current work is to propose and
capture the effect of such cell level changes through x-ray implement a fully automated CAD system for early breast
imagery. Screening with mammography is an effective tool cancer detection from mammography. The suggested system
to reduce the mortality rates, in which the images has an is intended to be economical system with reasonable
indication to what happening in cell level. There are variant complexity to be appropriate for early inspection method in
examination systems that may be different from country to our low income countries. The proposed CAD system is
country, also other aspects such as examination period, age, founded on PSO and utilizing the support vector machine
methods, technologies, and regulation diverge extremely [4]. (SVM) for classifying the region of interest, which is
For example, some countries recommended that two abstracted from mammogram images, into three classes (i.e.
radiologists read the mammograms with one of them is normal, malignant and benign). PSO is used to determine the

XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE

optimum number of clustered groups while performing the The ROI automatically was extract from the region in the
segmentation, by which there is no need for prior step to center of the breast near nipple point. In order to distinguish
decide the number of clustered groups. PSO is recommended the extracted ROIs, an eight texture features for each ROIs of
to be utilized in the field of medical images due to its different sizes are calculated. The final step used the SVM
uncomplicatedness and its great performance. classifier with the extracted feature vectors to classify ROIs
into the three density classes. The used SVM is based on two
The paper is put in as follows: section II lists a review for different kernel functions (i.e. Gaussian, polynomial and
the latest developed CAD systems. Section III presents the Radial Basis Function (RBF)). The method, which was
mathematical background of the utilized algorithms in the evaluated using 322 images from Mini-MIAS dataset, had
proposed system's building blocks. Section IV demystifies achieved overall accuracy of 89.7%.
the details of the proposed system (i.e. the design steps and
the implementation details of each block). Section V clarifies On the other hand, [14] presents a new system for breast
the conducted experiment, while section VI demonstrates the density classification in mammogram images. The breast
results, explained discussion and comparisons with recent region were extracted automatically from mammogram
published researches. Section VII concludes and abstracts the image background by applying a combined technique
achieved objectives and advises a future effort. between the thresholding technique and edge flag algorithm.
A new segmentation method for pectoral muscle extraction
II. LITERATURE SURVEY
333302/ from breast region using K-means clustering
Through the past two and a half decades, many technique and based on intensity value pixel in dense region.
researchers have studied and proposed various methods for In this system the segmentation step performed by using
implementing CAD system which is concerning the breast thresholding technique to segment the pixels in the digital
cancer detection in the mammograms. [11] proposed two mammogram into two separate class dense and fatty tissue.
computer-aided diagnosis systems for breast tissues density Mammogram density can be estimated by calculate the
recognition and discriminates it and compared between them. percent fatty and percent dense in the mammographic. The
One of these systems is founded on PCA (i.e. principal algorithm has been evaluated on a set of 180 mammogram
component analysis) and the SVM. On the other hand, the images from MIAS dataset. The samples include 60 dense,
other system is based on PCA and the Probabilistic neural 60 glandular and 60 fatty images. Percentages of fatty and
network (PNN) PNN. Firstly in that system, the ROIs are dense area which are calculated in previous step were used as
extracted manually from mammogram images in size input to the decision tree classifier to classify the
200×200 pixels and then five texture features are obtained. mammogram images into dense, glandular and fatty. Their
The interested five features are the mean, standard deviation, proposed system achieved 98% classification accuracy for
kurtosis, entropy and skweenes, which are derived from breast density classification.
Laws’ masks. After obtained the texture feature vectors of the
ROIs, PCA is applied for reducing the dimension of the III. MATERIALS AND METHODS
feature space. Finally, for classifying the breast density into
dense and fatty, the texture features is used as input to the Particle Swarm Optimization (PSO)
SVM and PNN classifiers to accomplish this task. The The social behavior of swarms of insects, birds or fish
accuracy obtained when using PNN classifier with four PCA inspires PSO to simulate their actions while they dynamically
component is 92.5%. These PCA components calculated over move and communicate [15]. PSO is stochastic search
texture features which are computed using Laws’ masks and technique and based on population. It relays on the motion of
assuming the length equals 7. The classification accuracy is the interacted swarm groups. PSO algorithm utilizes the
enhanced and achieved 94.4% when the SVM classifier is theory of social interaction to resolve complex life problem.
also applied for the first four PCA components. In this case, It simulates the concept of social mind behavior in a manner
the PCA components are acquired by the texture features that integrates self and social experience with one another.
calculated with length 5-Laws’ masks. Moreover, PSO is simpler than other meta-heuristic
algorithms, because it involves only the description of the
[12] proposed a new system for characterizing the density
problem and a limited parameters in order to solve that
for breast tissue from mammograms images for categorize
problem [16]. PSO utilizes a number of individual named
the ROI into the three classes of breast tissue mentioned
particles that compose a hovering swarm, at a velocity ,
before. In this method, the ROI was manually extracted from
looking for the optimal solution over a multi dimensional
the bottom half of any mammogram image and excepting the
searching space. Each point in the searching space is a
pectoral muscle and the background region. Then the
particle. Each particle adjusts its velocity depending on self
statistical features were extracted from the ROI. The texture
flying experience and the neighbor particles' flying
feature vector is provided as input to K-nearest neighbor
experience as well. In other words, the PSO system joins
classification (K-NN) process concatenated with 3-various
global search techniques with local search techniques, aiming
distance metrics (i.e. Euclidean, cosine and town-block).
to equilibrium the exploitation with exploration. That
These distances are individually and integrated between them
explains why PSO is used it in the recognition of the anomaly
and applied for the classifier. This algorithm was tested on
which is demanding the knowledge about the local and global
322 images from mini-MIAS dataset and achieved a
information. [17, 18]. Each particle have to memorize its
classification accuracy of 91.16%. Moreover, [13] proposed
position that results in its best result. The result refers to the
an effective pre-processing method for mammogram images
solution obtained by the particle and also known as fitness
and breast density classification. In their method, they
(cost function). The best solution (fitness) achieved by a
remove the noise, labels and other artifact from mammogram
particle up to now be pointed out as personal best ( ). On
images by applied thresholding technique and morphological
the other hand, the best solution (fitness) achieved by all PSO
operations. The pectoral muscle area was removed from
particles up to now be pointed out as global best ( ). The
breast region by applying polynomial curve fitting method.
particles require assistance in discovering where to complete Area: The area is the number of pixels in a shape.
their search. They interchange the obtained information so far
by other neighboring particles. That explain why the PSO A = Σ , (X) (1)
algorithm tracking the . Compactness: Compactness is representing the ratio that
Textural Features clarifies of the area (A) of an object to the area of a circle
with the same perimeter ( ).
The systematic local variation computed over the pixels’
intensities is describing the Image texture. Image texture = (2)
analyses is applied in order to calculate the image features
needed in any pattern recognition application. Before Circularity: Circularity describes how close a shape of
computing this feature the image pixels must be grouped into object should be to a circle.
similar regions. For the breast cancer classification = (3)
applications, the texture features prove its usefulness in
distinguish masses from normal tissues. Therefore, a set of Brightness: In general malignant masses’ intensities are
texture feature is extracted from the mammograms and greater than those benign masses’ intensities.
utilized by the dedicated classifier in order to discriminate ( ) (4)
between abnormal and normal pattern [19]. Eccentricity: Eccentricity represents the ratio between the
Texture descriptors are resulting from numerous methods short axis’s lengths to the long axis’s length of an object
such as; firstly, the first order statistic computed over pixel’s
intensity. Secondly, using wavelet transforms to represent the = 1− (5)
image sub-band and multi-scale. Thirdly, Laws measures.
Fourthly, statistics of pixel intensities, second order Convexity: the relative amount that an object differs from
histogram, or using GLCM (i.e. Grey Level Co-occurrence a convex object.
Matrix).
= (6)
GLCM features Elongatedness: The ratio between the width and length
One of the most magnificent texture analyses' methods is of the region bounding rectangle.
GLCMs. CLCM is defined as the square matrix that has a
dimension and holds the number of occurence of a = ( ∗ )
(7)
combination from a gray level values. Where is the
number of the gray combined levels. For gray scale images, Support Vector Machines
the GLCM descripes the properties of the spatial distribution SVM classifiers accompanying with machine learning
calculated over the values of pixel’s intensity [20, 21]. algorithms. SVM is utilized in multi-class clustering
Equation (1) is the form of GLCM’s matrix for an input problems. SVM could separate objects that associated to
image; where ( , ) indicates the summation of pixel’s “ ” different classes by defining decision boundaries
occurrence in the definite spatial association to a pixel “ ”. (hyperplanes) [24]. The SVM composed of two stages which
The computation of the GLCM depends on a set of are the training and testing stage. At the training stage, a set
parameters as follow: of training data is prepared to be the input to that phase. Each
point in the training data is labeled with a certain class that it
1. which is the number of the combined gray levels: belongs to. These training data is denoted as training vector
usually this value is considered to be 256 grey levels. in working space. Obviously other, unknown data vector has
2. Pixels' distance between each other: taking into to follow the same distribution.
consideration that the matrix is computed over non-
neighbors pixels Vectors A
3. Angle: indicates a pair of pixels direction and takes a Margin
values of (0, 45, 90, 135)
, , … ,

, , … , Optimal
(1)
⋮ ⋮ ⋮ ⋮ hyper plane
, , … ,

Shape Features
It is very important to differentiate malignant from benign Vectors B
tumor. Several primary shape features are considered as a key
feature for this task. In general, shapes descriptors (i.e. shape Fig. 1: SVM functions
features) are several set of numbers that are produced to Figure 1 illustrate the idea of SVM for building the
describe a given object shape [22]. Shape features can be hyperplans. This idea behind SVM have to retain as much
divided into two classes, region features and boundary safety margin, which points toward how far the decision
features [23]. The common used shape features (descriptors) boundary is from the data. In working space the data points
include the Area, compactness, circularity, Eccentricity, can be linearly separable or multi-dimensionally separable.
brightness, Elongatedness and the Convexity. These The SVM kernel function varies according to the described
descriptors is selected to be applied on the segmented tumor hyperplane (i.e. linear or nonlinear) and it is described by a
to accomplish the task of calculating shape features at the dot product in the space of higher dimension.
feature extraction step of the proposed method.
 IV. PROPOSED METHOD
A CAD system (computer-Aided detection) is proposed
for breast cancer in mammograms. The suggested CAD
system consists of five steps as shown in Figure 2. First, the
image preprocessing step which removes the labels and
artifacts. Second is the pectoral muscle removal from the
breast area. Third, the segmentation in which the region of
interest (ROI) is separated from breast image based on the
particle swarm optimization algorithm Fourth, interested
features are extracted that both the texture and shape features
are computed over ROI. Fifth, the classification step based on
multi-class SVM is conducted in order to divide the detected
ROIs into normal and abnormal (benign and malignant). The
detailed description of the suggested CAD system is
described in the following.

Fig. 3: Breast anatomy illustrates the pectoral muscle region in

mammography [26].

Pectorial
Preprocessing Muscule Segmentation Classification
Removal

Figure 2: CAD System using Particle Swarm Optimization algorithm Fig. 4: Different types of Breast mass in mammograms

Preprocessing Segmentation
Mammograms define the intensity of the breast density in The process of finding set of pixels that correlated
the image, where the fatty tissue appear as dark region and together is called “segmentation”. The segmentation process
glandular tissue appear as light region on the images. Hence, aims to decompose the digital mammogram image into
locating the mass represent a very hard task in the regions, which have symmetrical characteristics. Each
mammogram images. Therefore, it is important to improve identified regions represents the information that it belongs to
the fineness of the image which requires preprocessing of the and structuring elements to distinguish the ROIs. The main
targeted image before analysis and diagnosis. aim of segmentation the mammogram images in this CAD
The preprocessing is a necessary step for the system is region of interest segmented from the breast tissue
mammography images in order to accomplish the task of as shown in Figure 4. Image segmentation is the second step
noise removal, artifacts and label removal and background of breast cancer detection using CAD schemes, which
removal. To fully preparing the mammogram image for the separating suspicious regions that may contain masses from
segmentation step, the pectoral muscle has to be removed out breast tissue, i.e., partitioning the breast mammogram image
of the breast region. Scanning mammograms generate into several regions, and extracting ROIs and suspicious mass
artifacts (i.e. scanning artifacts, high and low intensity labels candidates from the mammogram image.in our proposed
and tape artifacts), as in the case of the Mini-MIAS dataset System , We chose the metaheuristic algorithm Particle
images. For this purpose we select automatically only the Swarm Optimization (PSO).to perform this task
area containing the breast region and removing these Proposed Fitness (cost) function for Segmentation
artifacts. For Artifacts and noise removal step we using an
automatically threshold suitable for Dense categories for each For the sake of segmenting the mammography images, a
mammogram image and then keep just the largest area which fitness function is designed in such a way to compromise
includes the breast with the pectoral muscle region. between two desired parameters to be optimized. These
parameters are first; the summation of distances ( ) between
Pectoral Removal all pixels and each other of the same cluster given by the
The upper right or left corner of the mammogram image following equation 8. The second parameter is the number of
represent the location of pectoral muscle. The pectoral clusters K
muscle region represents a high density and it appears ( , ) ⋯ ( , )
brighter than breast tissue that could interfere the
=∑ ⋮ ⋱ ⋮ (8)
examination of mammograms as shows in Figure 3. In the
density mammogram the pectoral muscle region are not ( , ) ⋯ ( , )
visible and very difficult to be removed, In some cases, = (2)
pectoral muscles are visible and easy to extracted from the
breast tissue[14]. where is the total fitness function. , are a weight
factor which are fractions less than one and their sum is 1.
For this purpose we applied, a fully automatic pectoral The weight factors are used to combine the two objectives
muscle removal method for successful extract the pectoral into single function and limit the maximum of the function to
muscle region is proposed [25]. one.
 V. EXPERIMENT low intensity artifacts (label and other noise). Figure 5(b)
shows the breast region after selecting the large object,
Mini-Mias Database removing the label and noise and orienting the breast image.
For evaluated the proposed system, we used Digital Figure 5(c) represents the results of extracting the pectoral
mammography images which obtained from the mini-MIAS muscle area from mammogram.
dataset. This dataset consists of right and left breast images
of all breast density categories glandular, fatty and dense On the other hand, in figure 7 in Column (a) the original
breasts [27]. The Mini-MIAS Dataset consisting of three mammogram the second column of that figure illustrates the
hundred twenty two record of left and right breast images for output of the segmentation step shows the segmented ROI
one hundred sixty one patients. The mammogram images are using PSO. The first row illustrate an image of normal case,
categorized into three categories: malignant, benign and second row illustrate an image of benign case and last row
normal Divided into 208 normal cases, 63 benign cases and illustrate an image of malignant case.
51 malignant cases. The abnormalities consist of 5 categories The step of features extracting computes 20 different
as follows: architecturally irregular, Ill-defined, features used in classification. Figure 7 presents the value of
Circumscribed, asymmetrical and speculated masses. The “volume” feature for the different types of masses. It is clear
images’ size is 1024x1024pixels. The images are represented that there is a definite barrier between the three classes which
in grayscale in which the pixel intensity ranges between 0 could be determined by the chosen classifier. This barrier
and 255. A nonlinear multi-class SVM classifier is used for also clear in figure 8 which shows a map of 3 different
classify segmented ROIs in mammograms images into the features of the three classes. These features are standard
three classes. The chosen kernel for SVM is radial basis deviation, energy and homogeneity.
function (rbf) kernel which is the popular kernel among
researches for its high accuracy among other kernels. From
the total of 322 mammogram images from mini-MIAS
dataset, 207 are normal cases, 64 are benign cases and the
remaining 51 are malignant cases.
VI. RESULTS AND DISCUSSION
3B

Figure 5 illustrates the results from the first step of the

proposed system as discussed in section IV. In this step, the
breast area is identified and the other unwanted objects are
excluded.

Fig. 7: The volume feature for different classes

(a) (b) (c)

Fig. 5:( a) original image ;(b)breast region;(c) pectoral muscle removal.

Fig. 8: Three Feature plan for differenet classes

To validate the proposed system we used all the

mammogram images (322 images) from mini-MIAS dataset.
The extracted twenty features for all cases are used to train
and test the classifier. The dataset was randomly divided into
10 equal sized subsets (10 folds). The classifier is then
trained on 9 parts and evaluated on the tenth part. The
process is then repeated using each of ten subsets as a
validation set. This is represent the applying of cross-
validation. The performance is calculated as the average for
K folds results to find the overall value. As mentioned in
(a) (b) table 1 the proposed algorithm detects accurately 85.4%
Fig. 6: (a) original image (b) ROI segmented normal from abnormal which is better than the compared
method. The classifying benign from malignant with
The image in figure 5(a) shows an example of original accuracy of 89.5%, and gain 81% total accuracy achieved
mammogram from Mini MIAS database contains on high and when classifying three classes normal, benign and malignant.
 TABLE 1: COMPARISON WITH PREVIOUS METHODS breast cancer," European journal of radiology, vol. 82, pp. 417-423,
Previous Method [28] Proposed Method 2013.
-PSO [8] P. Taylor, J. Champness, R. Given-Wilson, K. Johnston, and H. Potts,
Segmentation -Fuzzy interference system PSO "Impact of computer-aided detection prompts on the sensitivity and
based edge detection: specificity of screening mammography," 2005.
-Texture feature -Texture feature [9] J. D. Keen, J. M. Keen, and J. E. Keen, "Utilization of computer-aided
Features detection for digital screening mammography in the United States,
-Shape features -Shape features
classification method SVM Non linear SVM 2008 to 2016," Journal of the American College of Radiology, vol. 15,
2-class Normal based pp. 44-48, 2018.
83.8% 85.4%
Accuracy [10] P. Luo, W. Qian, and P. Romilly, "CAD-Aided Mammogram
2-class Abnormality Training1," Academic radiology, vol. 12, pp. 1039-1048, 2005.
_ 89.5%
based Accuracy [11] J. Virmani, N. Dey, and V. Kumar, "PCA-PNN and PCA-SVM based
3-class based CAD systems for breast density classification," in Applications of
_ 81%
Accuracy intelligent optimization in biology and medicine, ed: Springer, 2016,
pp. 159-180.
VII. CONCLUSION [12] K. Vaidehi and T. Subashini, "Breast tissue characterization using
The presented work proposes a Computer Aided combined K-NN classifier," Indian Journal of Science and
Technology, vol. 8, p. 23, 2015.
Detection system (CAD) for breast cancer that detects the
[13] S. S. Devi and S. Vidivelli, "Classification of breast tissue density in
ROIs in digital mammograms. The segmentation of breast digital mammograms," in 2017 International Conference on
tissue process in this proposed system is implemented in one Innovations in Information, Embedded and Communication Systems
step by using the PSO. The proposed system was evaluated (ICIIECS), 2017, pp. 1-7.
on mammogram images from mini-MIAS dataset. The [14] R. Nithya and B. Santhi, "Comparative study on feature extraction
images were preprocessed in order to remove labels, artifacts method for breast cancer classification," Journal of Theoretical and
and pectoral muscle region from the breast image for Applied Information Technology, vol. 33, pp. 220-226, 2011.
performing the segmentation process efficiently. After [15] S. Ait-Aoudia, E.-H. Guerrout, and R. Mahiou, "Medical image
segmentation using particle swarm optimization," in 2014 18th
extracted the ROI using PSO algorithm, we have computed International Conference on Information Visualisation, 2014, pp. 287-
the texture and shape features. The nonlinear support vector 291.
machine is used for classification of the ROI into three [16] N. Jain, U. Nangia, and J. Jain, "A review of particle swarm
classes; normal, benign and malignant tissue. The design and optimization," Journal of The Institution of Engineers (India): Series
implemented of the proposed CAD system is fully automated B, vol. 99, pp. 407-411, 2018.
including the preprocessed and pectoral muscle removal [17] P. Ghamisi, M. S. Couceiro, F. M. Martins, and J. A. Benediktsson,
which performed automatically. Besides, at the segmentation "Multilevel image segmentation based on fractional-order Darwinian
particle swarm optimization," IEEE Transactions on Geoscience and
step, the number of clusters is chosen automatically. Hence, Remote sensing, vol. 52, pp. 2382-2394, 2013.
there is no need for the intervention of a radiologist during [18] N. G. Raju and P. N. Rao, "Particle swarm optimization methods for
any stage. The CAD system is compared with recently image segmentation applied in mammography," Int. Journal of
methods and results are satisfy compromise between Engineering Research and Application, vol. 3, pp. 1572-1579, 2013.
efficiency, cost and speed as well. The particle swarm [19] A. Humeau-Heurtier, "Texture feature extraction methods: A survey,"
optimization algorithm, which is borrowed into the field of IEEE Access, vol. 7, pp. 8975-9000, 2019.
segmenting the mammogram images, is recommended to be [20] A. Makandar and B. Halalli, "Classification of mass type based on
utilized in the field of medical images due to its segmentation techniques with support vector machine model for
diagnosis of breast cancer," in 2017 International Conference on Data
uncomplicatedness and its great performance. The proposed Management, Analytics and Innovation (ICDMAI), 2017, pp. 81-86.
system is intended to be economical system with reasonable [21] A. M. Santos, A. O. de Carvalho Filho, A. C. Silva, A. C. de Paiva, R.
complexity to be appropriate for early inspection method in A. Nunes, and M. Gattass, "Automatic detection of small lung nodules
our low income countries. For future work the PSO is in 3D CT data using Gaussian mixture models, Tsallis entropy and
promising tool that could be hybridized with any other SVM," Engineering applications of artificial intelligence, vol. 36, pp.
27-39, 2014.
machine learning algorithm to optimize the overall
[22] A. M. Anter and A. E. Hassenian, "Computer aided diagnosis system
performance. for mammogram abnormality," in Medical Imaging in Clinical
REFERENCES Applications, ed: Springer, 2016, pp. 175-191.
[23] J. Shabbir and T. Anwer, "Artificial Intelligence and its role in near
[1] National Cancer Registry Program of Egypt. Available: future," arXiv preprint arXiv:1804.01396, 2018.
http://www.cancerregistry.gov.eg
[24] M. Abdel-Nasser, H. A. Rashwan, D. Puig, and A. Moreno, "Analysis
[2] A. S. Ibrahim, H. M. Khaled, N. N. Mikhail, H. Baraka, and H. of tissue abnormality and breast density in mammographic images
Kamel, "Cancer Incidence in Egypt: Results of the National using a uniform local directional pattern," Expert Systems with
Population-Based Cancer Registry Program," Journal of Cancer Applications, vol. 42, pp. 9499-9511, 2015.
Epidemiology, vol. 2014, p. 18, 2014.
[25] N. El-Sokary, A. Arafa, A. Asad, and H. Hefny, "Computer-Aided
[3] C. Di Maggio, "State of the art of current modalities for the diagnosis Detection System for Breast Cancer Based on GMM and SVM," Arab
of breast lesions," in Breast Cancer, ed: Springer, 2008, pp. 99-126. Journal of Nuclear Sciences and Applications, vol. 52, pp. 142-50,
[4] K. K. Jacobsen, L. Abraham, D. S. Buist, R. A. Hubbard, E. S. 2019.
O’Meara, B. L. Sprague, et al., "Comparison of cumulative false- [26] N. Saidin, H. A. M. Sakim, U. K. Ngah, and I. L. Shuaib,
positive risk of screening mammography in the United States and "Segmentation of breast regions in mammogram based on density: a
Denmark," Cancer epidemiology, vol. 39, pp. 656-663, 2015. review," arXiv preprint arXiv:1209.5494, 2012.
[5] E. Azavedo, S. Zackrisson, I. Mejàre, and M. H. Arnlind, "Is single [27] J. Suckling, J. Parker, D. Dance, S. Astley, I. Hutt, C. Boggis, et al.,
reading with computer-aided detection (CAD) as good as double "Mammographic Image Analysis Society (MIAS) database v1. 21,"
reading in mammography screening? A systematic review," BMC 2015.
medical imaging, vol. 12, p. 22, 2012.
[28] K. B. Soulami, M. N. Saidi, and A. Tamtaoui, "A cad system for the
[6] R. L. Birdwell, "The preponderance of evidence supports computer- detection of abnormalities in the mammograms using the
aided detection for screening mammography," Radiology, vol. 253, metaheuristic algorithm particle swarm optimization (pso)," in
pp. 9-16, 2009. International Symposium on Ubiquitous Networking, 2016, pp. 505-
[7] C. Dromain, B. Boyer, R. Ferre, S. Canale, S. Delaloge, and C. 517.
Balleyguier, "Computed-aided diagnosis (CAD) in the detection of

View publication stats