IRJET- Comparison of Breast Cancer Detection using Probabilistic Neural Network and Support Vector Machine

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 06 | June 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 3045
Comparison of Breast Cancer Detection using Probabilistic Neural
Network and Support Vector Machine
Shalinia G1, Dr. T.N.R Kumar2
1Student, Dept. of CSE, MSRIT, Karnataka, India
2Associate Professor, Dept. of CSE, MSRIT, Karnataka, India
---------------------------------------------------------------------***---------------------------------------------------------------------
ABSTRACT - Cancer is one of the deadliest diseases that the
human beings are encountered. Of all the types of cancers
found the breast cancer is the most common disease that has
highest death rates in women. Thetraditionalapproachesthat
are used to diagnose a cancer mainly depends upon the
doctor’s experience, investigation and visualization. Theerror
rate is high because human eye can’t recognizethepatternsso
easily therefore there is need for accurate diagnosis of
mammogram images. The main objective of this work is to
develop an automated system which identifies a tumour in
digital mammograms images. For this purpose, machine
learning algorithms can be used. Certainadaptivefiltersalong
with segmentation needs to be done before theclassificationis
done. Neural networks will aid in the training process and
then the prediction can be done appropriately. The tumor are
classified into the following classes benign, malign and
normal. By this classification process it can be found if the
cancer cells will spread or not. This will also help in increasing
the efficiency of the process of identification of breast cancer.
Keywords: Image Classification, Segmentation, GLCM,PNN,
SVM.
1. INTRODUCTION
The medical field keeps on improvising in ways by which
they can fight several diseases and this field has seen
exponential unidirectional growth. Majority of research
focus has been towards developing ways to counter deadly
diseases such as cancer. A disease where a set off cells tend
to concentrate into one particular region thereby leading to
formation of lump or tumour is termed to be as cancer. Out
of ninety million, the count of those that have been infected
by cancer has been twelve million which accountstocloseto
fifteen percent of the total count. The major general types of
cancer that are found are lung cancer, colon cancer, breast
cancer, pancreatic cancer, prostate cancer, ovarian cancer,
brain cancer. Breast tumour ismajorcausesfordeathcaused
due to these cancerous cells especially amongst women.
According the report availablebyWorldHealthOrganization
of the International Agency for Research on Cancer (IARC),
out of the eighteen million women that were identified who
are suffering from breast cancer nine million patients pass
away. Lung, breast and colorectumcancersare deemedto be
the ones that are more frequent, contributing around eleven
percent of the over-all cancer occurrence problem.
There are four stages in cancer namely stage 1, stage 2 a, b
and c, stage 3 and stage 4. Stage 1 contains small tumor and
spreads slowly in the formation of lymph can causing
infections. Stage 2a as well as 2b consists of always larger
than 2cm it is not part of lymph that was spread in stage 1.
There is a large tumor caused separately. In stage 2c cancer
will be due to lymph nodes then can spread along with
connecting tissues. The lymph spreads along with muscles
and may spread till bones when it is stage 3. In stage 4 the
tumor will spread in all body parts such as brain and liver.
Therefore, there is a need to decrease the impact caused by
cancer. When it comes to reducing the impact there are two
ways by which it can be achieved. One way is finding new
innovative ways to diagnose the disease thereby reducing
the treatment time. In second method, the tumour can be
detected at an earlier stage thereby reducing the damage
caused on the body. In this project we are formulating a
method by which we can detect the lumps or tumor by using
machine learning techniques and we are restricting the
project to breast cancer. Breastcancerwaschosenbecauseit
is common and more people suffer due to this disease, the
more the people suffer the greater is the need to find a
solution. Machine Learning techniques can be utilized
iteratively to find a solution to identify the existence of any
cancerous lump present in any body part with a greater
extent of accuracy and reliability.
2. LITERATURE SURVEY
Cancer is a disorder, the cure should be the major objective
through technical examination. Disease curing at initial
stages can be useful. Many techniques from the literature
are accessible for detecting cancer. Researchers have
discovered and implemented many ideas related to
detecting of cancers. The works of literature mainlydiscuss
current cancer recognition methods. Numerous fields and
ideas have applied in discovery of cancer. The foremost
fields in recognizing the technique incorporate image
processing, neural networks, and, etc. Mammogram image
system is being used since very long time and researches
have been carrying out.
2.1. Image Enhancement
It is a method which is provoked from realitythattumors be
inclined to be optimistic than the neighboring as specified

by Tang et al. (2009) [7]. Simple awareness is to utilize
approaches to increase the disparity of tumor areas. This
approach has developed to increase the value and
mammograms readability.
2.1.1. Direct contrast enhancement methods
The researchers in 2014, illustrated applications of
morphology mathematics processes in the area of digital
image processing. In the continuation, it presents as well as
discusses about the latest algorithm depending on
arithmetical morphology operators, can capable to identify
the micro calcification in mammograms images [8].
Applications of few operators to shape an algorithmapplied
to digital images. Rebuilding with isotropic organizing
element using erosion operation, followed by contrast
improvement in segmentation to discover novel dominant
algorithms. It is more effective to identify contrast
enrichment and also the micro calcification was benign or
malignant in mammography imagery. They have then used
added algorithm to obtain minimal areas from the image.
2.1.2. Wavelet enhancement methods
The researchers selected the move towards for finding
malignancy in a mammogram image. As it improvedtheSNR
portion of the lesion being identified and eradicated the
false-positive conclusion [9]. Thus an algorithm is proposed
for mammograms improvement that had refining intention
segmentation of dissimilar mammograms structures. The
enrichment method utilized wavelet decomposition,
restoration, morphological methods,local scaling.Following
improvement, separating areas was performed and from
each of the segmented regions, a group of features
computed. A ranking system is used for classification.
The researchers in 2009, it deals with two problems and
solution for it: Detection of tumor as doubtful areas with an
extremely feeble dissimilarity to the surroundings and
feature extractions that classify tumor [10]. The tumor
identification method works as (a) enhancement of
Mammogram (b) tumor part segmentation (c) feature
extraction by segmented tumor part (d) usage of SVM
classifier. Image excellence and understandable level is
improved. The mammogram improvement method
comprises of top hat action, discrete wavelet transform and
filtering. Contrast stretching gives rise for improving image
contrast. Mammogram segmentation of image plays a main
part for improving recognition, analysis of breast cancer.
Popular segmentation process made use is thresholding.
From segmented breast area, the features are removed.
Later phase classifies the area by using the SVMclassifier.75
images were tested.
Balakumaran et al. (2010), they have projected analgorithm
for spotting microcalcification in a mammogram. The
suggested microcalcification exposure comprises
mammogram superiority development utilizing multi
resolution investigation [11]. It can be feasible to identify
modular segments like micro calcification precisely by
presenting shape data. The efficiency of algorithm for
microcalcification identification was definite by
investigational results.
2.2 Region based algorithm
Region growing algorithm Bocchi et al. (2008),declaredthat
the micro calcification is primitive indications of breast
cancer. Here a method is proposed to identifyandcategorize
microcalcification. So as to determine microcalcification
bunches, specific attention to the exploration of noticed
lesions. To define the mammographic image a fractal model
is used [12]. It allows the usage of a coordinated filtering
phase to improve micro calcification in opposition to the
framework. An algorithm, linked by a neural classifier, finds
out existing lesions. Consequently, another fractal model is
made use of to investigate their collection.
Shanmugavadivu et al. (2014), projected a Wavelet
Transformation which makes use of morphological pre-
processing, area properties and seed area emerging to take
away the digitization noises, for removing the inner tissue
and to overcome radiopaque artifact, therefore segmenting
irregular masses perfectly [13].Thejoinedpossibleofregion
growing and wavelet benefits to activemassessegmentation
which guarantees for the worth of projected method.
2.2.1. Region of Interest Method
Alolfe et al. (2008), confirmed that collections of micro
calcification are premature indications of cancer that tends
for improving a CAD framework which aids for the
radiologist in analyzing micro calcifications patterns in
computerized mammograms before and quicker than
regular selection programs [14]. The suggested approach is
performed in 3 phases: (a) the region of interest choosing of
32×32 pixels size, (b) feature taking out step depends on
wavelet decomposition for locally preprocessed image to
figure the relevant each bunch features and (c) the
categorization stage that classifies among normal and micro
calcification patterns. In the this stage, 4 techniques are
used, K-Nearest Neighbor classifier (K-NN), fuzzy classifier,
Neural Network (NN) classifier and Support vector machine
(SVM) classifier.
Mohd Khuzi et al. (2009), stated that digital mammogram,
effective technique for early breast detection modality. It is
directly stored in a computer to grow automatic framework.
Digital image processing approaches used to increase the
images quality [15]. Extra data consists of a masses location
centers and masses radius. The taking out from features of
texture about ROIs is achieved by applying grey level co-
occurrence matrices (GLCM). The result shows GLCM @ 0,
45, 90 and 135 degree,which provideimportanttexturedata
toward classify among masses.

Rabi Naraiyan Pandya et al. (2009), says usual concurrent
choice for early discovery of breast tumor in women.
Although, the view of thephysicianhasa conspicuousimpact
on the mammogram elucidation.Theadvancedstudyaims to
improve an image processing algorithm for identification of
micro calcification also mass lesions to support early
exposure of breast tumour [16]. Work suggested here deals
with the drawing out features like microcalcification also
mammograms masslesionsforearlyon recognitionof breast
tumour. The advanced procedure is depended on a 3-step
process: (a) RegionsofInterest(ROI)projectedrequirement,
(b) Extraction of feature based on OTSU thresholding (c)
two-dimensional wavelet transformation. Research
recommended exposing of micro calcification and mass
tumorfrommammogrampicturesegmentation, examination
were tested on a number of images from mini-MIAS. The
algorithm implementation was using Matlab and it executes
effectively on computer, a digital mammogram as gathered
information for evaluation.
2.3 Probabilistic Neural Network
Al-Timemy et al (2009) mentionedaboutthearrangementof
threatening and kind breast tumor relied upon Fine Needle
Aspiration Cytology (FNAC) and Probabilistic Neural
Network (PNN). This data set includes 30 features where
input layer is represented to PNN. Massspectrometry-based
proteomics provides an accomplished approach for perfect
analysis of various disorders. However, there are issues in
the “mass spectral data” like complexity of, noise presence
that makes the examination of proteomic model extremely
complex.
Neural network-depended framework isprojectedby Xuu et
al (2009) for proteomic pattern examination. This method
chiefly includes 3 phases namely selection of feature on the
basis of statistical test, categorization from RBFNN a PNN,
lastly results in optimization over ROC study.
Ankita Satyendra Singh (2017), in this paper, designed and
obtain the outcomes of breast masses categorization usinga
data set of 322 images as well as considering 20 malignant
and 20 benign constrained masses, gambled, architectural
misrepresentation and 20 usual mammograms.
Subsequently the ROI abstraction, respectively mass is
characterized with 22 features of texture. Previously
categorization, choosing of feature was accomplished by 60
ROIs. The PNN classifier made use of 60 imagestocategories
over PNN classifier, Accuracy, Specificity, Sensitivity are
accomplished.
2.4 Support Vector Machine Learning methods
An innovative mechanism for recognition of clusteredmicro
calcification in mammogram are recommended by
Xinnsheing Zhag et al (2009). Correct recognition of micro
calcifications is important reason in CAD outline. Hence to
improve the presentation of detection, a Bagging then
Boosting depended Twin Support Vector Machine (BB-
TWSVM) is proposed to detect micro calcifications. The
process contains 3 parts particularly pre-processing of
image, extraction of feature and also theBB-TWSVMpart. At
first, straightforward physical object elimination filter anda
well-made high-pass filter are utilized for every micro
calcification preprocessing. Then collective pictures
extractors of feature are utilized to extort 164 features of
image. Within the collective area ofimagefeatures,themicro
calcifications identification process is projected as a
supervised learning and categorization drawback.
Fathma Edaoudi and Regragui Fakhita (2011) declared
cancer representation of the foremost oftenanalyzedcancer
in ladies. So as to diminish death, primitive identification of
breast cancer is significant; as a result of identification is a
lot of doubtless to achievesuccesswithintheprimaryphases
of sickness, given a novel methodology for automatic
recognition of clustered micro calcificationsincomputerized
mammograms. This innovative methodology makes use of
the textures coding of mammographic pictures on idea of
that Haralick options are calculated for SVM classification
intention. With the help of examination outcomes those got
within the literature, strategy of coding is tried which
tremendously decreases the estimating time of Haralick
vector parameters released. In advance, rates of
categorization originated by victimization,thecodedimages
enhanced a lot.
Xeen Zhang Sheng (2014) projected to categorize and
discover Mcs. Classification issues are developed as
distributed feature study dependent mostly categorization.
The visual data-rich terminology of training samples is
physically created bya collectionofsamplesthat incorporate
Mcs elements. From the earlier Mcs position truth in
mammograms, distributed feature learningisassimilatedby
the regular fact of Mcs in mammograms, distributed
characteristic learning is obtained by spare feature study
dependent mostly Mcs categorization algorithmic rule by
using Twin Support Vector machine (TWSVM). Projected
technique is applied 2 DDSM dataset and then equated with
SVMs for exploring its performance with identical dataset.
Tests depicts that performance of projected technique is
further economical, higher than state-of-art strategies.
3. PROPOSED METHODOLOGY
This section deals with the collection of MRImammographic
images and different algorithm used for identification and
recognition oftumors.Pictureunitsofdistinctmammograms
picture utilized in the examination were accumulated bythe
Mammographic Image Analysis Society (MIAS) is an
association of UK investigation organizations engaged with
the acknowledgmentofmammogramsadditionallyhasmade
a database of computerized mammograms. The database
holds 322 digitized mammograms also been decreased to a
200-micron pixel frame and the images are 1024x1024.

A community that consists of a sum of 321 mammographic
images, 206 signify normal images, 51 malignant including
62 benign images. The models in the MIAS database are
collected in the Portable Gray Map setup. Eachimageis8-bit
grey level range images among 256 various grey levels (0–
255).In this analysis utilized 20 benign, 20 normal also 20
malignant mammograms which are thick, fat and fatty-
glandular moreover have irregularities like surrounded
asymmetry, masses,, ill-defined masses moreover all the
mammograms are changed toward .JPG format. The below
figure shows the 40 malignant which is a type of cancerous.
The cells are anomalous and distribute randomly also 40
benign and 40 normal which a type is of the non-cancerous
group.
3.1. Preprocessing
Preprocessing is measured as important step so as to find
the orientation of the mammogram images and to enhance
the standard of the image. The breast images collected will
be subjected to preprocessing for noise removal using
adaptive median filteringtechniquesothattheresultismore
suitable than the original image for a specific application.
The adaptive Median Filter classifies picture elements as
noise by comparing every pixel within the image to its close
neighboring pixels. The dimensions of the neighborhood is
adjustable, because the threshold. A picture element that's
completely different from a majority of its neighbor, which
are not structurally aligned with those pixels to that it's
similar, is labelled as impulse noise. These noise pixels are
then replaced by the median pixel value of the pixels within
the neighborhood that have passed the noise labelling test.
3.2 Extracting Region of Interest
The challenging part for image analysis is extracting ROI
from image. There are several classical types of
segmentation process is applied to extract ROI. The
identified regions are analyzed to detect various types of
defects which has covered over a mammography images. In
suspicious abnormality the ROI is extracted and cropping
operation is applied. By referring the abnormal area,theROI
take approximate pixel range of circle surrounding the
abnormal region. Hence, the extracted ROI is free from the
background information and noises.
3.3 Gray-Level Co-occurrence Matrix (GLCM)
The transition between two pixels of gray level is used in
gray-level co-occurrence matrix to extract the texture of the
image. It also provide joint distributionofneighboringpixels
of gray level pairs within an image. For classification of the
breast malignancy a descriptorsfromthematrixisextracted.
For the computation of GLCM, established the spatial
relationship between two pixels i.e. reference pixel and
neighbor pixel so it form a different combination of gray
pixel values in an image. Let the element of the GLCM for a
given image be q (i, j) of size M x N containing gray levels G
ranging from 0 to G-1. Hence, the element q (i, j) is defined
as:
……………. (1)
Where the locations of reference pixel and its neighboring
pixel are (x, y) and (x+∆x, y+∆y) respectively. And q (i, j |∆x,
∆y) is relative frequency of two pixels in a given
neighborhood of distance (∆x, ∆y) having gray level values.
In an image, the two pixel with intensities i and j separated
between two neighboring resolution cells with a distance D.
And the direction of neighboring pixel w.r.t the pixel
reference denoted by θ.
The extracted texture features are:
•Energy: It is the sum of the square values in the GLCM.
….……… (2)

•Contrast: It measure the variation in the matrix
distribution, a local change in an image in a number,
imitating the image clarity and shadow depth of texture.
…………… (3)
•Entropy: It measure the randomness of an image texture
and the minimum value is achieved when co-occurrence
matrix is equal for all the values.
………….. (4)
•Correlation Coefficient: It measure the joint probability
existence of the quantified pixel pairs.
…………… (5)
•Homogeneity: It measure the distribution of closeness
element in the GLCM to the GLCM diagonal.
……………… (6)
3.4. Probabilistic Neural Network
PNN is made up of 3 layers of nodes including input, where
activation function are hidden and output.Ininput,theinput
data is set which is nonlinearly transformed into a high
dimensional hidden layer by the basic functions then linear
transformation of the hidden output layer.
The below figure shows the architecture for PNN that
classifies K equal to two classes,howeveritcan beprolonged
to any number of K classes. On the left side the input layer
comprises N nodes of the N input feature vector. In the
hidden layer nodes, receives thewiderangeofinputfeatures
X. And in the output layer node for each feature vector a
Gaussian function is grouped to feed their functional values.
Therefore k output nodes will be presented.
Figure 3.2: PNN Architecture.
For the class K, Gaussian values arebeensummedandscaled
so that the probability of the sum function is unityandforms
probability density function. Assume there are P example
feature vectors {x: p =1… P} labelled as class 1 and Q (p)
example features vectors {y: r=1… R} labelled as class 2.
Therefore, in the class 1 represent P nodes and in the class 2
represent R nodes. Each Gaussian is centered on the
particular class 1 and class 2 points x and y i.e., feature
vectors as shown in the equation.
…………… (7)
The output node k is sum of the values established after the
hidden nodes in the k group called Gaussian mixed model.
The sums are defined by
………….. (8)
Where input feature vector is denoted by x, σ1 and σ2 are
the standard deviations for Gaussianinclass1andclass2,the
dimension of input vectors denoted by N, the number of
center vector in class1 denoted by P, the number of center
vector in class2 denoted by R, the centers of class1 and 2
respectively denoted by x (p) and y(r) and the Euclidean
distance between x and x (p) is denoted by ||x-x(p)||.
The maximum value of f1(x) and f2(x) will decides which
classes it belong to. There is no computing weights or
iteration for each class may be it reduced by thinning which
are close to one another by making σ larger. Here the PNN
recognize three different classes namely benign, malignant
and normal based on feature database.
4. RESULT & DISCUSSION
We have used Matlab, R2014a, 64-bit for the experimental
set up. The RGB image of size 256x256 is considered for
experimental purpose. The comparison between SVM and
PNN classifier is done respectively.

Figure 4.1: Display the different stages when the
program is executed.
The above figure indicates format of the resulting image
display. Firstly input image should be uploaded. After image
is been uploaded Adaptive Median Filter, Segmentation,and
classifier for the input image and class will be identified and
final processed and resulting image will be displayed on the
right hand side.
Figure 4.2: Selecting of input image from the database.
The above figure depicts the uploadingoftheimagefromthe
computer for processing and classification of the type of
disease, it will be done after processing using the algorithm
or the code.
(a) (b) (c)
Figure 4.3: a. Benign input image b. Preprocessed
image c. Segmented image.
Figure 4.4: This shows the performance evaluation of
PNN & SVM.
The above figure depicts the image after performing
segmentation, classification through using PNN and SVM. The
class of the image will be identified after processing the image
through all the three steps i.e. adaptive median filtering,
Segmentation and classifier. So the result is obtained which is
same as the predicted one.
Figure 4.5: Indicates the uploading of malign image for
processing and identifying the class.
The above figure depicts the uploading of the known example
malign class cancer image and verifying the same image and
finding out the class through different steps of processing.
(a) (b) (c)
Figure 4.6: a. Malign input image b. Preprocessed Image
c. Segmented image.

Figure 4.7: Indicates the image undergoing classifier
block for processing and identification of the cancer
class.
The above figure depicts the image after processing in the
segmentation block, classification is done through the 2
types that is PNN and SVM, steps are carried out throughthe
two classifier processes also and cancer class will be
identified for the uploaded image.
Benign Malign Normal
Benign 38 1 1
Malign 2 36 2
Normal 1 2 37
Table 4.1: Confusion Matrix
Finally from the 120 images was considered which included
40 images of benign, 40 images of malignant and 40 images
of normal. After classification 38 images of benign images
were properly classified, 36 images from malignant images
were appropriately classified and 37 images from normal
images were correctly classified as shown in the table 4.1.
PNN SVM
Accuracy 92.5 90.83
Sensitivity 90.47 85.71
Specificity 93.58 93.58
Precision 88.37 87.80
Recall 90.47 85.71
F-Measure 89.41 86.74
G-Mean 92.01 89.56
Table 4.2: Performance measurement of the PNN and
SVM
Compared the classifier between PNN and SVM. Accuracy,
Sensitivity and so on obtained by PNN was higher than SVM,
which clearly depicts that applying PNN would give higher
accuracy than SVM. Finally PNN is used as a classifier to
distinguish mammogram and classify it into normal or
malignant class.
Figure 4.8: Comparison graph between PNN and SVM.
The Performance of the PNN and SVM is as shown in the
above figure some the parameter such as Sensitivity
obtained by applying PNN was 90.47 and SVM was 86.71.
Precision obtained by using PNN was 88.37 and SVM was
87.80. Precision obtained by PNN was higher than SVM.
Accuracy obtained by using PNN was 92.5 and SVM was
90.83 which clearly depicts that applying PNN would give
higher accuracy than SVM.
5. FUTURE SCOPE
Detection of breast cancer is a field in which there is a
constant scope for enhancement. The detection process can
be done using several other detection methods and new
image processing tools can be used.
Classification and prediction can be done using new
decision-makingalgorithms. Multiplealgorithmscanbeused
to generate more accurate results and at the same time the
speed of detection can also be increased.
The performance measurement parameters need to be
maximized as much as possible, this will ensure that the
detection is of higher quality and the results given while
detection can become more reliable.
The dataset can be altered and more appropriateimagescan
be added. The better the quality of the dataset thebetterwill
be the performance parameters. All the images added must
be unique and of better detailing in order to produce to
desired result.
6. CONCLUSION
In this proposed work, it has become evident it is possible to
predict the presence of tumor and at the same time it can
classify if the tumor present is benign or malign. This will
also help in diagnosis to identify, if the cancer will spread to
other parts of the body. The classification is done using
Probabilistic Neural Network and based on the network
formed the training process is also done. Theclassificationis
done after the completion of two vital processes. Firstly, the

Adaptive Median Filter is applied to the image followed by
which segmentation needs to be done.
Training is done using a set of one hundred and twenty
images which comprises of three classes namely: normal,
benign and malign. Each class is built up with forty images.
Testing can be done with any randomimagethatwill fall into
any ne of this class and that image will be classified into the
right class. In case there are more images that need to be
added to the base dataset it can be done by performing the
Probabilistic Neural Network training to the entire set of
images. Performance measure is also calculated and
displayed in the user interface that was created solely for
this purpose. And in the classification for the benign images
38 images of benign were properly classified, 36 images
from malignant images were appropriately classifiedand37
images from normal images were correctly classified. There
is also a graphical representation of the performance
measure. The performance measure does include the
following parameters that are displayed: accuracy,
sensitivity, specificity, precision, recall, g-mean, f-measure.
This software like setup can be used to aid in detection of
cancer. It can help in reducing certain painful acts such as
biopsy and can make the detection process much simpler. It
will help prevent mental stress that the patients with tumor
or lumps will undergo.
7. REFERENCES
[1] “Latest world cancer statistics, International Agency for
Research on Cancer (IARC) press releases 2013, World
Health Organization (WHO),”
http://www.iarc.fr/en/media/centre/pr/2013/index.pp.
[2] “Globocan project 2012, International Agency for
Research on Cancer (IARC), World Health Organization:
cancer fact sheets,” http://globocan.iarc.fr.
[3] “Statistics of breast cancer in India,”
http://www.breastcancerindia.net/statistics/ stat
global.html.
[4] S. V. Engeland, “Detection of mass lesions in
mammograms by using multiple views.” 2006.
[5] L. Tabar and P. Dean, “Mammography and breast cancer:
the new era,” International Journal of Gynecology &
Obstetrics, vol. 82, no. 3, pp. 319 – 326, 2003.
[6] H. Cheng, X. Shi, R. Min, L. Hu, X. Cai, and H. Du,
“Approaches for automated detection and classification of
masses in mammograms,” Pattern Recognition,vol.39,no. 4,
pp. 646 – 668, 2006.
[7] Tang J, Rangayyan R.M, Xu J, El Naqa I and Yang Y (2009),
“Computer-Aided Detection and Diagnosis of Breast Cancer
with Mammography: Recent Advances”, IEEE Transactions
on Information Technology in Biomedicine, Vol.13,No.2,pp.
236- 251.
[8] Mohamed Lagzouli and Youssfi Elkettani (2014), “A New
Morphology Algorithm for Microcalcifications Detection in
Fuzzy Mammograms Images, International Journal of
Engineering Research & Technology”, Vol. 3, No.1, pp.729 -
733.
[9] Dominguez A.R and Nandi A.K (2007), “Enhanced Multi
level thresholding segmentation and Rank based Region
selection for detection of masses in mammograms”, IEEE
International Conference on Acoustics, Speech Signal
Processing, pp. 440-452.
[10] Ireaneus Anna Rejani Y and Thamarai Selvi S (2009),
“Early Detection of Breast Cancer Using SVM Classifier
Technique, International Journal on Computer Science and
Engineering”, Vol 1,No.3, pp.127-130.
[11] Balakumaran T and ILA Vennila (2011), “A Computer
Aided diagnosis system for micro-calcification cluster
detection in digital mammogram”, International Journal of
Computer Applications, Vol.34, No.1, pp.39-45.
[12] Bocchi L, Coppini G, Nori J and ValliG(2008),“Detection
of single and clustered micro-calcificationsinmammograms
using fractals models and neural networks”, Medical
Engineering & Physics, Vol.26, pp.303-312.
[13] Shanmugavadivu P, Sivakumar V and Suhanya J (2014),
“Wavelet Transformation Based Detection of Masses in
Digital Mammograms, International Journal of Research in
Engineering and Technology”,Vol.3, No.2, pp.131-138.
[14] Alolfe M.A, Youssef A.M, Kadah Y.M and Mohamed A.S
(2008), “Computer-Aided diagnostic system based on
wavelet analysis for micro-calcification detection in digital
mammograms, Proceedings of the IEEE Cairo International
Biomedical Engineering Conference (CIBEC 2008)”, 978-1-
4244-2695-9/08.
[15] Mohd. Khuzi A, Besar R, Wan Zaki W, and Ahmad N
(2009), “Identification of masses in digital mammogram
using gray level co-occurrence matrices”, Biomedical
Imaging Interv Journal, Vol.5, No.3, pp.1-13.
[16] Rabi Narayan Panda, Bijay Ketan Panigrahi and Manas
Ranjan Patro (2009), “Feature extraction forclassification of
micro-calcifications and mass lesions in mammograms”,
International Journal of Computer Science and Network
Security, Vol.9, No.5, pp. 255-265.
[17] Dubey R.B, Hanmandlu M and Gupta S.K (2010), “Level
set detected masses in digital mammograms, Indian Journal
of Science and Technology”, Vol.3, No.1, pp.9- 13.
[18] Mario Mustra, Mislav Grgic and Kresimir Delac (2012),
“Enhancement of micro-calcifications in digital

mammograms, intelligent image features extraction in
knowledge discovery systems”, IWSSIP, pp.248-251.
[19] Dr.D.Devakumari and V.Punithavathi (2018),
“Comparison of noise removal filters for breast cancer
detection in mammogram images”, International Journal of
Pure and Applied Mathematics Volume 119 No. 18 2018,
3863-3874. [20] R. Ramani, Dr. N.Suthanthira Vanitha andS.
Valarmathy (2013), “The Pre-Processing Techniques for
Breast Cancer Detection in Mammography Images”, I.J.
Image, Graphics and Signal Processing, 2013, 5, 47-54.
[21] Subhash Chandra Bose J, MarcusKarnanandSivakumar
R (2010), “Detection of masses in digital mammograms”,
International Journal of Computer and Network Security,
Vol.2, No.2, pp.78-86.
[22] Nalini Singh, Ambarish G Mohapatra and Gurukalyan
Kanungo (2011), “Breast Cancer Mass Detection in
Mammograms using K-means and Fuzzy C-means
clustering”, International Journal of Computer Applications.
Vol. 22, No.2, pp.15-21.
[23] Bouyahia S, Mbainaibeye J and Ellouze N (2009),
“Wavelet based micro-calcifications detection in digitized
mammograms, ICGST-GVIP Journal”, Vol.8, No.5, pp.23-31.
[24] Nizar Ben Hamad, Khaled Taouil andMedSalimBouhlel
(2013), “Mammographic micro-calcificationsdetectionusing
discrete wavelet transform”, International Journal of
Computer Applications, Vol.64, No.21, pp.17-22.
[25] Xinsheng Zhang, Xinbo Gao and Minghu Wang (2009),
“MCs detection approach using Bagging and Boosting based
twin support vector machine”, IEEE International
Conference on Systems, Man and Cybernetics (SMC 2009),
pp.5000-5505.
[26] Fatima Eddaoudi and Fakhita Regragui
(2011),”Microcalcifications detection in mammographic
images using texture coding, Applied Mathematical
Sciences”, Vol.5, No.8, pp.381- 393.
[27] Xin-Sheng Zhang (2014), “A New Approach for
Clustered MCs Classification with Sparse Features Learning
and TWSVM”, The Scientific World Journal, Article ID
970287, pp.1-8.
[28] Al-Timemy A.H, Al-Naima F.M, and Qaeeb N.H (2009),
“Probabilistic neural network for breast biopsy
classification”, MASAUM Journal of Computing, Vol.1, No.2,
pp.199-205.
[29] Xu Q, Mohamed S.S, Salama M.M.A, Kamel M and
Rizkalla K (2009), “Mass spectrometry-based proteomic
pattern analysis for prostate cancer detection using neural
networks with statistical significance test-based feature
selection”,IEEETorontoInternational ConferenceonScience
and Technology for Humanity (TICSTH), pp.837- 842.
[30] Ankita Satyendra Singh (2017), “Mass Classification of
Mammogram Images using Selected Textural Features with
PNN Classifier”, International Journal ofInnovativeResearch
in Computer and Communication Engineering, An ISO3297:
2007 Certified Organization Vol.5, Special Issue4,June2017
[31] Vasumathi, D., and Raju, G.T (2015), “Mammogram
Image Preprocessing for detection of masses in Breast
Cancer”. International Journal of Engineering and Computer
Science.
BIOGRAPHIES
Student in Computer Science
Department of Ramaiah Institute of
Technology. Areas of interest include
image processing and machine
learning.
Associate Professor in Computer
Science Department of Ramaiah
Institute of Technology. Areas of
interest include image processing,
software engineering and computer
networks.

IRJET- Comparison of Breast Cancer Detection using Probabilistic Neural Network and Support Vector Machine

More Related Content

What's hot (20)

Similar to IRJET- Comparison of Breast Cancer Detection using Probabilistic Neural Network and Support Vector Machine (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Comparison of Breast Cancer Detection using Probabilistic Neural Network and Support Vector Machine