Artificial Neural Networks in Image Processing for...

Review ArticleArtificial Neural Networks in Image Processing forEarly Detection of Breast Cancer

M. M. Mehdy,1 P. Y. Ng,1 E. F. Shair,2 N. I. Md Saleh,3 and C. Gomes2

1Department of Computer and Communication System Engineering, Universiti Putra Malaysia, Serdang, Selangor, Malaysia2Department of Electrical and Electronics Engineering, Universiti Putra Malaysia, Serdang, Selangor, Malaysia3Department of Chemical and Environmental Engineering, Universiti Putra Malaysia, Serdang, Selangor, Malaysia

Correspondence should be addressed to C. Gomes; [email protected]

Received 15 January 2017; Accepted 9 March 2017; Published 3 April 2017

Academic Editor: Po-Hsiang Tsui

Copyright © 2017 M. M. Mehdy et al. This is an open access article distributed under the Creative Commons Attribution License,which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Medical imaging techniques have widely been in use in the diagnosis and detection of breast cancer.The drawback of applying thesetechniques is the large time consumption in the manual diagnosis of each image pattern by a professional radiologist. Automatedclassifiers could substantially upgrade the diagnosis process, in terms of both accuracy and time requirement by distinguishingbenign and malignant patterns automatically. Neural network (NN) plays an important role in this respect, especially in theapplication of breast cancer detection. Despite the large number of publications that describe the utilization of NN in variousmedical techniques, only a few reviews are available that guide the development of these algorithms to enhance the detectiontechniques with respect to specificity and sensitivity. The purpose of this review is to analyze the contents of recently publishedliterature with special attention to techniques and states of the art of NN in medical imaging. We discuss the usage of NN in fourdifferent medical imaging applications to show that NN is not restricted to few areas of medicine. Types of NN used, along withthe various types of feeding data, have been reviewed. We also address hybrid NN adaptation in breast cancer detection.

1. Introduction

Breast cancer is one of the main causes of death amongwomen and the most frequently diagnosed non-skin cancerin women [1]. Breast cancer occurs when the cell tissuesof the breast become abnormal and uncontrollably divided.These abnormal cells form large lump of tissues, whichconsequently becomes a tumor [2]. Such disorders couldsuccessfully be treated if they are detected early. Thus, it isof importance to have appropriate methods for screening theearliest signs of breast cancer.

Microcalcifications and masses are the earliest signs ofbreast cancer which can only be detected using modern tech-niques. Microcalcifications are clusters of calcium depositswhich are very small in size and present inside the soft breasttissues [2]. Generally, detection of masses in breast tissues ismore challenging compared to the detection of microcalcifi-cations, not only due to the large variation in size and shapebut also because masses often exhibit poor image contrast

when usingmammography [3].The difficulty in classificationof benign and malignant microcalcifications also causes asignificant problem in medical image processing.

Automated classifiers may be useful for radiologists indistinguishing between benign andmalignant patterns.Thus,in this paper, an artificial neural network (ANN) which canbe served as an automated classifier is investigated. Inmedicalimage processing, ANNs have been applied to a variety ofdata-classification and pattern recognition tasks and becomea promising classification tool in breast cancer [4]. ANNapplications in mammography, ultrasound, and MRI and IRimaging for early detection of breast cancer are reviewed inthis paper.

Image features can be distinguished inmany aspects, suchas texture, color, shape, and spatial relations. They can reflectthe subtle variance inmany degrees.Thus, different selectionsof image features will result in different classification deci-sions. These classifications can be divided into three types:first, the method based on statistics, such as Support Vector

HindawiComputational and Mathematical Methods in MedicineVolume 2017, Article ID 2610628, 15 pageshttps://doi.org/10.1155/2017/2610628

https://doi.org/10.1155/2017/2610628

2 Computational and Mathematical Methods in Medicine

Machine; second, the method based on rule, such as decisiontree and rough sets; and third, artificial neural network [5].

In the early 1980s, there was an increment in the use ofneural networks in the field of image and signal process-ing. The main benefit was the reduction in manipulationtime due to the parallel-distributed processing behaviorof neural networks [6]. Then the network had been usedwidely in the common image processing methods such asvector quantization, eigenvector extraction, 2D pulse codemodulation, or 2D filtering [7]. The artificial neural networkresembles the function of the biological neuron, and it iscomposed of neurons with different layers and these neuronsare interconnected by numeric weights; these weights canbe changed due to the learning behavior of the network toapproach the optimum result. Usually in image processingapplications, the number of the neurons is directly related tothe number of pixels in the input image [8], and the numberof layers depends on the processing steps.

For cancer detection and classification, image segmen-tation has been widely used. Many image segmentationmethods, based on histogram features, edge detection, regiongrowing, or pixel classification, have been trained usingANNs [9].

Although the technology related to ANN in breast cancerdetection has rapidly moved forward during the last fewyears, there is a dearth of critical literature review on thesubject which is a distinct drawback for further developmentof the technologies. This paper is an attempt to fulfill thatvacuum in the field of image processing in the early detectionof breast cancer.

2. Applications of ANNs

2.1.Mammogram. Mammography is one of themost effectivemethods used in hospitals and clinics for early detection ofbreast cancer. It has been proven effective to reduce mortalityas much as by 30% [3]. The main objective of screeningmammography is to early detect the cancerous tumor andremove it before the establishment of metastases [3, 10, 11].The early signs for breast cancer are masses and microcal-cification but the abnormalities and normal breast tissuesare often difficult to be differentiated due to their subtleappearance and ambiguous margins [3]. Only about 3% ofthe required information are revealed during a mammogramwhere a part of suspicious region is covered with vesselsand normal tissues. This situation may cause the radiologistsdifficult to identify a cancerous tumor. Thus, computer-aided diagnosis (CAD) has been developed to overcomethe limitation of mammogram and assists the radiologiststo read the mammograms much better [10]. ANN modelis the most commonly used in CAD for mammographyinterpretation and biopsy decision making. There are twoways used in ANN to assist in mammography interpretation:first, applying classifier directly to the region of interest (ROI)image data and second, understanding the situation from thefeatures extracted from the preprocessed image signals [12].Figure 1 shows an example of ANN structure with multi-featured input data and multi-hidden layers [12].

Microcalcification is deposition calcium in the soft breasttissues. They are quite minute in quantity and size. It isfound in a cluster or pattern of circles/lines together withextra cell activity in breast region [2]. Many researchershave developed CAD system using artificial neural networkto detect microcalcification. In early 90’s, research doneby Dhawan et al. [13] has defined image structure featuresusing the second-order grey-level statistics.The classificationwas based on implementing perceptron based 3-layer neuralnetwork and the network uses backpropagation algorithmin training which has been used successfully for a numberof pattern recognition and classification applications [13, 14].The entropy feature has significant discriminating power forclassification [13]. The group of researchers further extendedtheir research in investigating the potential of using second-order histogram textural features for their correlation withmalignancy. Several architectures of neural networks wereproposed to analyze the features extracted from segmentedcalcifications and it shows that the neural network gives goodresults for the classification of hard-to-diagnoses cases ofmammographic microcalcification into benign and malig-nant categories using the selected set of features [15].

Image segmentation is a technique used in image process-ing. Basically, segmentation is performed on the raw image todetect small, local, and bright spots. Research done by Kevinet al. [16] has drawn significant attention on the segmenta-tion process and neural network used. After segmentationprocess, ANN is performed to distinguish the segmentedobjects called candidates, as either microcalcifications ornonmicrocalcifications. The accuracy of the ANN is testedby having a set of labelled test images for determination oftrue positive (TP) and false positive (FP) detection rates.This ANN is using cascade correlation (CC) for patternclassification. It is a self-organizing ANN which runs asupervised learning algorithm.TheCCANNapproach showsa promising result to detect microcalcification [16].

Not only image segmentation but also image registrationtechniques can be used for the breast cancer detection whereANN is performed to enhance the effectiveness of the cancerdetection. In Saini and Vijay [17], Grey-Level CooccurrenceMatrix (GLCM) features are extracted and used as input totrain artificial neural network based breast cancer detectionsystem. After that, the extracted features of known andunknown mammogram images have been compared usingfeed-forward backpropagation and Cascade forward back-propagation ANN to distinguish the malignant and benignimages. Feed-forward backpropagation network has highaccuracy of 87.5% compared to Cascade forward backpropa-gation network with 67.8% after optimizing the number ofneurons and number of layers [17].

In late 90’s, the application of ANN in CADmammogra-phy was found to have limitation in terms of data overfitting.Thus, Bayesian belief network (BBN) was compared withANN classification method to identify the positive massregions based on a set of computed features in CAD. Thesame database was used in ANN and a BNN with topologiesoptimization using a genetic algorithm (GA) to test theperformance and robustness of the ANN and BBN. However,the result shows that there is no significant difference between

Computational and Mathematical Methods in Medicine 3

Input layer(patient’s data)

Hidden layers Output layer(malignancy probability)

Input Node 1:calc.

calc.

distribution

calc. number

distribution

Input Node 2:

Input Node 3:

Input Node 4:

Input Node 5:

mass density

mass margin

age

Transfer function

Transfer function

Transfer function

Hidden layer 1Neuron 1

Hidden layer 2

Neuron 2

Hidden layer 3

Neuron 3

Hidden layer 3

Neuron R

Hidden layer K

Neuron 1

Hidden layer K

Neuron T

Output(probability ofbeing malignant)

···

···

···

∑

∑

∑

Hidden layer K

Neuron 2

∑

∑

∑

∑

Input Node R:

WeightsWeights

Weights

Figure 1: Structure of a typical ANN for classification of breast tumors in mammography [12].

using an ANN and using a BBN in CAD for mass detectionif the network is optimized properly [18]. In Alayliogh andAghdasi [11], wavelet-based image enhancement techniquehas been used to improve the detection of breast cancer. Inputfeature vectors containing spatial and spectral image wereemployed in neural network classifier. Microcalcificationdetection scheme and wavelet image enhancement havebeen investigated. Microcalcification detection has been per-formed by using amultistage algorithm comprising the imagesegmentation and pattern recognition to classify the micro-calcifications whereas biorthogonal spline wavelets have beenused in image enhancement to separate the image intofrequency bands without affecting the spatial locality. Theresult shows that spatial and spectral feature are promisingways to detect microcalcification [11].

Besides microcalcification, masses are the most impor-tant symptoms which are difficult to be detected and distin-guished accurately. A new algorithm based on two ANNs

(artificial neural networks) was proposed to detect thesemasses automatically. ANFIS and multilayer perceptron(MLP) classifier have been used for adjustment and filtration.Suitable methods and parameters should be applied to gethigh detection precision and low false positive (FP) [10]. Thedetection process was well adjusted and improved with thisproposed algorithm and the final diagnosis result showedthat the CAD scheme could simultaneously achieve compar-atively high detection precision and low false positive rate,even when the special masses are dealt with [10].

In mammography equipped with CAD system, the majorproblems developed are inconsistency and low classificationaccuracy. The accuracy can be improved by introducing anovel intelligent classifier which used texture information asinput for the classification of normal and abnormal tissues inmammograms. Dheeba et al. [3] used neutral network as anew artificial intelligent technique for the tissue classification.CAD system based on the optimized wavelet neural network


(a) (b) (c)

Figure 2: Results (from (a)–(c)): original image, image after first stage of NN processing, and image at second stage of NN processing usingGabor wavelets as input for mammogram image [20].

was designed and evaluated using Particle Swarm Opti-mization approach (PSOWNN). Optimization using heuris-tic algorithm is done to find appropriate hidden neurons,momentum constant, and learning rate during the trainingprocess. Thus, it will improve the classification accuracy inbreast cancer detection by reducing the misclassification rate[3]. In Zhang et al. [19], backpropagation neural network(BPNN) is introduced for the classification of benign andmalignant. The digitized mammogram used fuzzy detectionalgorithm to detect the microcalcification and suspiciousarea. BPNN gives a very promising result with 83.3% for theclassification [19].

Any defect in breast image obtained frommammogram ishighly advantageous to be detected automatically. In Lashkari[20], Gabor wavelets and ANN are used to classify normaland abnormal tissues which could increase the accuracy andsave radiologist’s time (Figure 2). Gabor wavelets transformshave a good attribute in image processing and computervision. The result shows that this combination of neural net-works has a good potential with 97% accuracy on unknowncases [20].

2.2. Ultrasound. Neural network (NN) also plays its rolein ultrasound images in detecting breast cancer. We willfirst look into the capability of NN in determining andrecognizing a region wheremalignant and benign lesions canbe found. Buller [21] was one of the first who used neuralnetwork in breast cancer detection for ultrasound images.In his work, he separated the training process for benignand malignant cases by feeding the first system only withimages containing benign lesion and the other with imagescontaining only malignant lesion. He also introduces “spiderweb” topology which are able to produce two vectors thatare further used in the classification process. The first vector

represent the localized effects in a defined neighborhoodand the other represents the global attributes. The techniquebrings high advantages as the spider web topology is sensitiveto small area and hence provides better results in smallarea by putting more weights on the localized effects. Thistechnique can actually be improved taking into account theextra parameters of texture and shape. As we know, malignand benign lesion has slightly different texture and shape. InRuggierol et al. [22], the researchers implement the NN usingboth texture and shape parameters. Transition probabilitymatrix and Run length matrix on the texture parameterhave been used to quantify the homogeneity of images whileshape parameter shows the irregularity of border of lesion.Besides, three-layer NN consisting of input, output, andhidden neurons was used. They also execute the trainingprocess by leaving one out before training and use the left outas tester. From the result, texture implementation achieved avery good result on both solid and liquid lesion.

Classification is an important technique used widely todifferentiate cancerous and noncancerous breasts. Denserbreast has higher risk in having cancer. Knowing this, Sahineret al. [23] in their paper describe the importance of textureimages in classification of dense breast tissue. They alsointroduce convolution neural network classifier to replacethe backpropagation methods where the images are feddirectly into the network. To measure the coarseness oftexture, grey-level difference statistics and features are used,whereas spatial grey-level dependence features will be show-ing the element distribution. The strength of this methodis that no image of tumor is fed into the network. Besides,segmentation does not need to be performed beforehand;instead a threshold value will be used. The drawback is thehigh computational cost which in turn makes the techniqueprobably unsuitable for real-time operation.


As early as 1999,NN classifier has been usedwith autocor-relation features to classify the breast cancer as benign ormalignant. Chen et al. [24] introduced a 25-input node mul-tilayer feed-forward NN which consists of 24-dimensionalimage feature vector from the ultrasound image, togetherwith a predefined threshold of the input layer to classifythe cancer. The introduced system has a relatively highaccuracy in classifying malignancies and thus could helpinexperienced operators to diagnose complicated ultrasoundimages. One of the striking advantages of NN is that it couldbe further optimized by supplying larger set of ultrasoundimages as the NN is well-trainable.

Self-organizing map (SOM) model is one type of neuralnetworks that can be trained without supervision; it iswidely used as a classifier in recent years. It expresses lowerdimensional data in a simple geometry compared with thecomplex high dimensional data such as that in Chen et al.[25]. SOM training is fairly easy as a desired output is notnecessary. SOM will automatically choose the closest inputvector. The classification of benign and malignant lesion isautomated.The only training needed is the data from textureanalysis. However, its accuracy is slightly lower than that ofthe multilayer feed-forward NN introduced by Chen et al.[24] previously. With a high negative predictive value of 98%,this CAD technique of SOM could potentially avoid benignbiopsies. In the same year, Chen et al. [26] came out with aHNN diagnostic system.The texture information features areextracted from four ultrasound images. The 2D normalizedautocorrelation matrix for the input is modified and 2-phaseHNN was used to combine the texture information from allof the ultrasound images.This study proves that the use of all4 images leads to amore promising result than the case whereimages are used separately.

Bootstrap is a statistical measure that relies on randomsampling with replacement. Combination of bootstrap tech-nique with neural network helps improve the accuracy. Chenet al. [27] implement bootstrap to perform random sampling;the observation was then recomputed. This technique isuseful where large amount of training data is not available, asit does not require much training data. However, the reducedamount of data should be compensated by adding imageanalysis component, in the bootstrap method.

The research done in 2002 used error backpropagationalgorithm to train the multilayer perceptron neural network(MLPNN) and resulted in an area index of the receiver oper-ating curve (ROC) of 0.9396 ± 0.0183 [28]. Seven morpho-logical features have been introduced to differentiate benignfrom malignant breast lesions with the use of MPNN [29].The morphological features were named as lobulation index(LI), elliptic-normalized skeleton (ENS), elliptic-normalizedcircumference (ENC), depth to width ratio (D :W), long axisto short axis (L : S), number of substantial protuberances anddepressions (NSPD), and the size of lesion. The MPLNNis also tested with different number of hidden neurons butall results lead to a similar performance. Accuracy of thetraining set and test set is better than SOM and MLP NNwith three inputs but on par with the accuracy of NN with25 autocorrelation features as inputs. Different inputs selected

and number of inputs may have an impact on the accuracy ofthe NN itself regardless of any types of NN techniques.

A year later, a research tested the NN by using only5 morphological features which are the spiculation, branchpattern, ellipsoid shape, brightness of nodule, and the numberof lobulations [30, 31]. Based on thesemorphological features,the difference of characteristics between the benign andmalignant could be seen as follows:

(i) Spiculation (benign: larger; malignant: smaller)(ii) Branch pattern (benign: fewer; malignant: more)(iii) Ellipsoid shape (benign: smaller; malignant: larger)(iv) Brightness of nodule (benign: larger; malignant:

smaller)(v) Number of lobulations (benign: fewer; malignant:

more)Latest research implements the hybrid method to

improve the conventional neural network method to detectthe malignancies of breast cancer. Combination of thek-means cluster algorithm with the backpropagation neuralnetwork (BPNN) is proven to provide an impressive perform-ance [32]. Figure 3 shows the result of image segmentationusing ANN to extract cysts from an ultrasound breast image[32].

2.3. Thermal Imaging. Thermal imaging has been used forearly identification of breast tumor and risk prediction sincethe 60s [33].Thermogram is a promising cutting edge screen-ing instrument as it can caution ladies of breast malignancyup to 10 years ahead of time [34]. Some studies have utilizedseveral types of ANNs to manipulate and classify IR images,by taking the IR image as an input to the ANN [35]. In2003, multispectral IR images were classified using LagrangeConstraint Neural Network (LCNN) which provides a betterdiagnosis for the physician [36]. Wavelet transformation isalso useful with ANN for multidimensional features of the IRimage, especially when it was found that the temperature ofthe breast is affected by many pathological factors includingthemental state [37]. Asymmetry discrimination between leftand right breasts can be done to produce statistical featuressuch as mean temperature and standard deviation that couldbe utilized as info parameters to a backpropagation ANN[33]. In 2007, thermographic image analysis was done byimplementing a special neural network that utilizes somefuzzy logic principles, called Complementary Learning FuzzyNeural Network (CLFNN). CLFNN takes many factors intoaccount such as family history and temperature difference ofthe statistical features between contralateral breasts [34]. Thesystem is widely used in several countries at present.

2.4. MRI. MRI technique has been used widely in medicalexaminations, especially for cancer investigation for fewdecades [38]. For the diagnosis to be done properly, breastregion should be extracted from other surrounding regionsand tissues using image segmentation methods [39]. Figure 4depicts such case as it was reported in the study [39].

Many neural networks models were utilized to aid MRIfor enhancing the detection and the classification of the


Figure 3: Segmentations of cysts for breast ultrasound image using ANN [32].

Figure 4: Multistate CNN used to segment small fatty breast and medium dense breast for MRI image [39].

breast tumors, which can be trained with previous casesthat are diagnosed by the clinicians correctly [40], or canmanipulate the signal intensity or the mass characteristics(margins, shape, size, and granularity) [41]. In 2012,multistatecellular neural networks (CNN) have been used inMR imagesegmentation to estimate the density of the breast regionsfor evaluation of the fat contents [39]. Hassanien et al.[38] introduced a hybrid model consisting of Pulse CoupleNeural Network (PCNN) and Support Vector Machines(SVM) to identify breast cancer from MR images. Anotherhybrid algorithm was presented by ElNawasany et al. in 2014by combining perceptron with the Scale Invariant FeatureTransform (SIFT) for the same purpose [42].

3. Discussion

For the last few decades, several computer-aided diagnosis(CAD) techniques have been developed in mammographicexamination of breast cancer to assist radiologist in overallimage analysis and to highlight suspicious areas that needfurther attention. It can help radiologist to find a tumorwhich cannot be spotted using naked eye. As technologieskeep growing, many researchers are concerned about thedevelopment of intelligent techniques which can be used inmammography to improve the classification accuracy. Thisartificial intelligence makes use of human skills in a moreefficient manner than the conventional mathematical modelsdo. Based on the research outcomes, ANN is proved to be agood classifier in mammography for classification of massesand microcalcifications. Implementation perceptron basedthree-layer neural network using backpropagation algorithmbecomes a pioneer in ANN mammography. Various ANNsdeveloped are based on the concept of increasing the truepositive (TP) detection rate and decreasing the false positive

(FP) and false negative (FN) detection rate for the optimumresult. Implementation of wavelet in ANNs such as ParticleSwarm Optimized Wavelet Neural Network (PSOWNN),biorthogonal spline wavelet ANN, second-order grey-levelANN, and Gabor wavelets ANN can improve the sensitivityand specificity which are acquired in masses and microcalci-fication detection.

For ultrasound applications, in the field of determiningbreast cancer malignancy, CAD frameworks utilizing ultra-sound images are widely used due to their nonradiationproperties, low cost, high availability, speedier results, andhigher accuracy. An improved version of the breast cancerdetection using ultrasound images has been introduced,whichworks on a three-dimensional ultrasound imaging thatcan give more in-depth information on the breast lesioncompared to the conventional two-dimensional imaging.This three-dimensional imaging joins each of the two-dimensional characteristics. Furthermore, in order to handlethe vulnerability nature of the ultrasound images, somemethods and methodologies based on ANN have also beenintroduced.Amajority of the researchworks that utilizeANNhave acquired noteworthy results. Hybrid methods, whichcombine two ANN techniques, have recently been developedfor the detection and classification of breast cancer. A two-phase hierarchical NN is also found to be promising ratherthan using the image analysis separately. It can also be seenthat the larger the number of inputs to theANN, the better theaccuracy of the output in identification and classification ofbreast cancer. However, the number of hidden neurons doesnot seem to have a big impact on the accuracy of the system.To state which individual ANN is the best is quite subjectivedepending on the application and various variables to beconsidered. Most of the ANN techniques for the ultrasoundapplication give good results in terms of accuracy, sensitivity,


Table1:Summaryof

metho

dswith

NNin

breastcancer

detection.

Stud

yMetho

dsInpu

tPu

rpose

Dataset

Classifi

erRe

sults

Dheebae

tal.

[3]

ParticleSw

arm

Optim

ized

Wavelet

NeuralN

etwork

(PSO

WNN)

Mam

mogram

Improvec

lassificatio

naccuracy

inbreastcancer

detectionandredu

cing

misc

lassificatio

nrate

216mam

mograms

PSOWNN

(i)Sensitivity94.16

7%(ii)S

pecificity

92.10

5%(iii)AU

C0.96853

(iv)Y

ouden’s

index0.86272

(v)M

isclassificatio

nrate0.063291

Xuetal.[10]

New

algorithm

basedon

two

ANNs

Mam

mogram

Classifi

catio

nof

masses

30casesa

nd60

mam

mograms

(con

taining78

masses)

ANFISandMLP

(i)True

positive(TP

)rate9

3.6%

(73/78),

(ii)N

umbero

fthe

FPsp

erim

age

0.63

(38/60).

Alay

liogh

andAgh

dasi

[11]

ANNand

biorthogon

alsplin

ewavele

tMam

mogram

Classifi

catio

nof

microcalcificatio

nclu

ster(MCC

)and

imagee

nhancement

40digitized

mam

mogram

ANN

(i)Sensitivity93%,

(ii)F

Prate(M

CC/im

age)0.82

Dhawan

etal.

[13]

(i)ANN

(ii)secon

d-order

gray-le

velstatistic

sMam

mogram

Classifi

catio

nof

significantand

benign

microcalcificatio

ns5im

ages

tructure

features

(i)Th

ree-lay

erperceptro

nbased

ANN

Thee

ntropy

featureh

assig

nificantd

iscrim

inatingpo

wer

forc

lassificatio

n

Chitree

tal.

[15]

ANN

Mam

mogram

Classifi

catio

nof

microcalcificatio

ninto

benign

andmalignant

(i)40

,60,and80

training

cases

(ii)151,131,and

111testcases

ANN

Neuraln

etworkisar

obust

classifier

ofac

ombinatio

nof

images

tructure

andbinary

features

into

benign

and

malignant


Table1:Con

tinued.

Stud

yMetho

dsInpu

tPu

rpose

Dataset

Classifi

erRe

sults

Kevinetal.

[16]

ANN

Mam

mogram

Classifi

catio

nof

microcalcificatio

nsand

nonm

icrocalcificatio

ns

24mam

mogramsw

itheach

containing

atleasto

neclu

stero

fmicrocalcificatio

ns

Cascadec

orrelatio

nANN

(CCANN)

(i)TP

detectionratefor

individu

almicrocalcificatio

nsis73%and

92%forn

onmicrocalcificatio

ns

Zhengetal.

[18]

ANNandBB

NMam

mogram

Com

pare

perfo

rmances

ofANNandBB

N3independ

entimaged

atabases

and38

features

ANNandBB

N

Perfo

rmance

level(𝐴𝑧value)

(i)ANN𝐴𝑧value

0.847±0.014(ii)B

BN𝐴𝑧value

0.845±0.011

(iii)Hybrid

classifier

(ANNand

BBN)

A zvalueincreased

to0.859±0.01

Zhangetal.

[19]

Digitize

mod

ule,

detectionmod

ule,

featuree

xtraction

mod

ule,neural

networkmod

ule,

andcla

ssificatio

nmod

ule

Mam

mogram

Classifi

catio

nof

microcalcificatio

nclu

sters/suspiciou

sareas

Fuzzydetectionalgorithm

(i)30

digitalimages

(15contain

benign

casesa

nd15

contain

malignant

cases)

Backprop

agationneuralnetwork

(BPN

N)

(i)Fu

zzydetectionrate(benign

84.10

%and80.30%

)(ii)C

lassificatio

nrates(feature

vector,𝑛=10is83.8%),(fe

ature

vector𝑛=14is72.2%)

Lashkari[20]

ANNandGabor

wavele

tsMam

mogram

Classifi

catio

nof

breast

tissues

tono

rmaland

abno

rmalcla

sses

automatically

(i)Im

ages

of50

norm

aland50

abno

rmalbreasttissues

(ii)6

5casesfor

training

setand

35casesfor

testing

set

ANNandGabor

wavelets

(i)Classifi

catio

nrate(te

sting

perfo

rmance

96.3%andtraining

perfo

rmance

97.5%)

Sainiand

Vijay[17]

Imager

egistratio

ntechniqu

eand

ANN

Mam

mogram

Classifi

catio

nof

benign

andmalignant

42mam

mogram

images

(30

benign

and12

malignant

images)

Feed-fo

rwardbackprop

agation

andCa

scadeforward

backprop

agationartifi

cialneural

network

Percentage

accuracy

(feed-fo

rward

backprop

agationnetworkis

87.5%andCascade

forw

ard

backprop

agationnetworkis

67.8%)


Table1:Con

tinued.

Stud

yMetho

dsInpu

tPu

rpose

Dataset

Classifi

erRe

sults

Bullere

tal.

[21]

Spider

web

topo

logy

with

NN

Ultrasou

ndClassifyandseparate

benign

andmalignant

lesio

n25

sono

gram

s(i)

NNcla

ssifier

(i)69%accuracy

inmalignant

(ii)6

6%accuracy

inbenign

(iii)66%accuracy

inno

lesio

ns

Ruggierolet

al.[22]

Texturea

ndshape

parameter

feeds

into

NN

Ultrasou

ndAu

tomated

recogn

ition

ofmalignant

lesio

n

(i)41

carcinom

as(ii)4

1fibroadeno

mas

(iii)41

cysts

(i)NNcla

ssifier

(i)95%accuracy

insolid

lesio

ns(ii)9

2.7%

accuracy

inliq

uid

lesio

ns

Sahinere

tal.

[23]

Con

volutio

nalN

Nwith

spatialand

textureimage

Mam

mogram

Classifi

catio

nof

mass

andno

rmalbreast

168mam

mograms

(i)Con

volutio

nNNcla

ssifier

(i)Av

eragetruep

ositive

fractio

nof

90%atfalse

positivefraction

of31%

Chen

etal.

[24]

Multilayer

feed-fo

rward

neuralnetwork

(MFN

N)

Ultrasou

ndClassifybenign

and

malignant

lesio

n140pathologicalproved

tumors

(52malignant,88benign

)MFN

N

(i)95%accuracy,98%

sensitivity

(ii)9

3%specificity

(iii)89%po

sitivep

redictivev

alue

(iv)9

9%negativ

epredictive

value

Chen

etal.

[25]

Self-organizing

map

(SOM)

Ultrasou

ndClassifi

catio

nof

benign

andmalignant

lesio

ns243tumors(82

malignant,161

benign

)SO

M

(i)Ac

curacy

of85.6,sensitivity

97.6%

(ii)S

pecificity

79.5%

(iii)Po

sitivep

redictivev

alue

70.8%

(iv)N

egativep

redictivev

alue

98.5%


Table1:Con

tinued.

Stud

yMetho

dsInpu

tPu

rpose

Dataset

Classifi

erRe

sults

Chen

etal.

[27]

Bootstrapwith

NN

Ultrasou

ndcla

ssificatio

nof

tumor

263sono

graphicimages

olid

breastno

dules

NN

(i)Ac

curacy

87.07%

,sensitivity

98.35%

(ii)S

pecificity

79.10

%(iii)Po

sitivep

redictivev

alue

81.46%

(iv)N

egativep

redictivev

alue

94.64%

Chen

etal.

[26]

2-ph

ase

Hierarchical

NeuralN

etwork

(HNN)

Ultrasou

ndDifferentiatebetween

benign

andmalignant

tumors

1020

images

(4different

rectangu

larregions

from

the2

orthogon

alplanes

ofeach

tumor)

HNN

4im

agea

nalyseso

feachtumor

appear

togive

morep

romising

resultthan

ifthey

areu

sed

separately

Chen

etal.

[28]

Wavele

ttransform

andneural

network

Ultrasou

ndDifferentia

ldiagn

osisof

breasttumorso

nsono

gram

s

242cases(161b

enign,

82malignant)

Multilayer

perceptro

nneural

network(M

LPNN)

(i)Re

ceiver

operating

characteris

tic(ROC)

area

index

is0.9396±0.0183

(ii)9

8.77%sensitivity,81.3

7%specificity

(iii)72.73%

positivep

redictive

value

(iv)9

9.24%

negativ

epredictive

value

Chen

etal.

[29]

Multilayer

feed-fo

rward

neuralnetwork

(MFN

N)

Ultrasou

ndDifferentiatebenign

from

malignant

breast

lesio

ns

1stset:160

lesio

ns2n

dset:111lesions

MFN

N(i)

98.2%training

accuracy

(ii)9

5.5%

testingaccuracy

Jooetal.[30]Artificialneural

network(A

NN)

Ultrasou

ndDeterminingwhether

abreastno

duleisbenign

ormalignant

584histo

logically

confi

rmed

cases(300benign

,284

malignant)

ANN

(i)100%

training

accuracy

(ii)9

1.4%testingset

(iii)92.3%sensitivity,90.7%

specificity


Table1:Con

tinued.

Stud

yMetho

dsInpu

tPu

rpose

Dataset

Classifi

erRe

sults

Jooetal.[31]

Digita

limage

processin

gand

artifi

cialneural

network

Ultrasou

ndDetermineb

reastn

odule

malignancy

584histo

logically

confi

rmed

cases(300benign

,284

malignant)

ANN

(i)91.4%accuracy,92.3%

sensitivity

(ii)9

0.7%

specificity

Zhengetal.

[32]

Hybrid

metho

d(unsup

ervised

k-means

cluste

r,supervise

dbackprop

agation

neuralnetwork

(BPN

N))

Ultrasou

ndClassifi

catio

nof

breast

tumorsa

sbenignor

malignant

125benign

tumors,110malignant

tumors

Com

binatio

nof

k-means

with

BPNN

(i)Re

cogn

ition

rate(94.5%

for

benign

,93.6%

form

alignant)

(ii)9

4%accuracy,94.5%

sensitivity

(iii)93.6%specificity

Foketal.[35]ANNwith

3Dfin

iteelem

ent

analysis

IRTu

mor

predictio

n200patie

nts

ANN

Goo

ddetection,

poor

sensitivity

Szuetal.[36]

Unsup

ervised

classificatio

nusing

Lagrange

Con

straint

Neural

Network(LCN

N)

Mid

andlong

IRim

ages

Early

detectionof

breast

cancer

One

patie

ntwith

DCI

SLC

NN

Bette

rsensitivity

Jaku

bowska

etal.[37]

ANNwith

wavelet

transfo

rmIR

Disc

riminationof

healthyandpathological

cases

30healthy

ANN

Accuracy

(%)

(fron

tal/side)

Raw:90/93,P

CA:90/93

LDA:93/97,N

DA:93/93

10with

recogn

ized

tumors

Accuracy

(%)

(fron

tal/side)

Raw:80/90,P

CA:80/90

LDA:90/90,N

DA:80/100


Table1:Con

tinued.

Stud

yMetho

dsInpu

tPu

rpose

Dataset

Classifi

erRe

sults

Koay

etal.

[33]

Backprop

agation

NN

IREa

rlydetectionof

breast

cancer

19patie

nts

Levenb

erg-Marqu

ardt

(LM)a

ndRe

silient

Backprop

agation(RP)

Accuracy

(%)

(RP/LM

)Who

le:

95/95

Quadrants:

95/10

0

Tanetal.[34]

Fuzzyadaptiv

elearning

control

networkfuzzy

neuralnetwork

IREa

rlydetectionof

breast

cancer

andtumor

classificatio

n

28healthy,43

benign

tumors,7

cancer

patie

nts

FALC

ON-AART

Cancer

detection(%

)(TH

/TDF)

Predicted:95,sensitivity

:100,

specificity:60

Breasttumor

detection(%

)(TH/TDF)

Predicted:84/71,sensitivity:

33/76,

specificity:91/6

2Breasttumor

classificatio

n(%

)(TH/TDF)

Predicted:88/84,sensitivity:

33/33,

specificity:95.5/91

Cardilloetal.

[40]

NNfora

utom

atic

analysisof

image

statistics

MRI

Early

detectionand

classificatio

n150exam

ssub

dividedinto

6grou

psby

contrast

NN

Bette

rinspecificity

Tzacheva

etal.[41]

Evaluatio

nof

signalintensityand

massp

ropertiesb

yNN

MRI

Automaticdiagno

sisof

tumors

14patie

nts

Feed-fo

rwardBP

NN

90%–100%sensitivity,91%

–100%

specificity,and

91%–100%

accuracy

Ertase

tal.

[39]

Extractio

nof

breastregion

sby

conventio

naland

multistateCN

Ns

MRI

Breastdensity

evaluatio

nandabno

rmality

localization

23wom

enCN

N

Averagep

recisio

n99.3±1.8

%True

positivev

olum

efraction

99.5±1.3

%False

positivev

olum

efraction0.1

±0.2%

Hassanien

etal.[38]

Image

classificatio

nusing

PCNNandSV

Mandusingwavele

tandfuzzysetsfor

enhancem

ent

MRI

Breastcancer

detection

70no

rmalcases,50

benign

and

maligncases

Hybrid

schemeo

fPCN

Nand

SVM

Accuracy

SVM:98%

Roug

hsets:

92%

ElNaw

asany

etal.[42]

Classifying

MR

images

byhybrid

perceptro

nNN

MRI

Early

detectionof

breast

cancer

138abno

rmaland143no

rmal

Perceptro

nwith

SIFT

Accuracy

86.74

%


specificity, positive predictive value, and negative predictivevalue. Another advantage of using the ANN in determiningbreast lesion is that the ANN can be trained to producebetter accuracy. Besides that, this ANN can be combinedtogether with not only another ANN technique but alsoother signal processing techniques such as wavelet to producebetter results.

For different techniques that have been utilized forthe breast cancer imaging, there are different methods ofdetection and classification according to the input parametersof that technique. For the IR imaging, it has been shownthat the detection of the breast cancer depends mainly onthe statistical features of the thermal behavior of the tissues(mean, standard deviation, etc.), as well as the asymmetrydifferentiation of the contralateral breasts. Therefore, imageclassification methods based on ANN are quite fruitful inthermography.

The MRI imaging is highly recognized as a reliabletechnique for tumor localization as well as early detectionand classification of cancer, as it is generally recommendedfor soft tissue recognition. Many image segmentation and 3Dextraction algorithms are applied in MRI applications, andrecently, many ANN classification types have been designedwith many fine specifications for MRI breast imaging.

A summary of methods with NN in breast cancer detec-tion has been given in tabulated form in Table 1.

4. Conclusion

Neural network plays an important role in detection ofcarcinogenic conditions in the breast. The technique acts asa stepping stone in the detection of cancer. In this review,we show that NN can be used in many medical applicationswhich we categorized into four main medical applicationsthat are widely used in breast cancer detection. These fourmedical applications include mammogram, ultrasound, andthermal and MRI imaging. This shows that NN is notrestricted by the application.

In all applications, NN’s main purposes were automatedclassification and segmentation. The types of data that needto be classified include calcification and noncalcification,benign and malignant, dense and normal breast, and tumor-ous and nontumorous. Neural network needs training data.Different types of data are fed into NN for training purposes.In early adaptation of NN, images of breast are being feddirectly into the NN. This method will perform well onlyif very large databases are available. In the case of usingsuch huge data, the concern was the storage, the time ofperformance, and the data availability. This flaw was realizedand being improved by taking the ROI into account, whichlowered the amount of dataset requirement tremendously.Researcher were then able to come out with better ideaswhere they now train the NN with feature vectors. In ourfindings, the features that can be used as training data includespiculation, branch pattern, shape, brightness of nodule,number of lobulations, margin, size of nodule, granularity,and texture.These features can be extractedmanually or usingimage analysis technique. Introducing of features did improve

the performance of NN in terms of size of training data andaccuracy.

Different variation of NN can be applied as classifier.Feed-forward backpropagation NN is by far the simplestform of NN, as the name suggest, the input nodes do nothave interrelation between each other, andmore importantly,the units do not form a repetitive cycle or loops. Feed-forward backpropagation can only pass data from currentlayer to subsequent layer; hence the data is moving inone fix direction from input to output. Cascade forwardNNs are somehow similar to feed-forward NNs; the onlydifference is that they include connections from not onlythe input, but also every previous layer to the subsequentlayers. ConvolutionNN is considered as a special type of feed-forward neural network where there are multiple layers ofsmall neuron collections that are able to process the portionsof input image.

The trend now is going towards hybrid NN like SOMmodel. Combination of statistical methods such as bootstrapis being used together with NN too. SOM and bootstrapmethods require lesser training data and hence are usefulwhen we do not have many training data. Besides, peopleutilize SVM with NN in order to achieve a better perfor-mance. In conclusion, NN is widely used in medical imageapplications, creatively combined with other methods inorder to achieve better accuracy, sensitivity, and also positivepredictive value.

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper.

Acknowledgments

This study has been supported by the Departments ofComputer and Communication Engineering, Electrical andElectronics Engineering and Chemical and EnvironmentalEngineering at Universiti Putra Malaysia (UPM).

References

[1] L. Hadjiiski, B. Sahiner, M. A. Helvie et al., “Breast masses:computer-aided diagnosis with serial mammograms,” Radiol-ogy, vol. 240, no. 2, pp. 343–356, 2006.

[2] T. Balakumaran, I. L. A. Vennila, and C. G. Shankar, “Detectionof microcalcification in mammograms using wavelet transformand fuzzy shell clustering,” International Journal of ComputerScience and Information Technology, vol. 7, no. 1, pp. 121–125,2010.

[3] J. Dheeba, N. Albert Singh, and S. Tamil Selvi, “Computer-aideddetection of breast cancer on mammograms: a swarm intelli-gence optimized wavelet neural network approach,” Journal ofBiomedical Informatics, vol. 49, pp. 45–52, 2014.

[4] S.-C. B. Lo, H.-P. Chan, J.-S. Lin, H. Li, M. T. Freedman, andS. K. Mun, “Artificial convolution neural network for medicalimage pattern recognition,” Neural Networks, vol. 8, no. 7-8, pp.1201–1214, 1995.


[5] C. Da, H. Zhang, and Y. Sang, “Brain CT image classificationwith deep neural networks,” in Proceedings of the 18th AsiaPacific Symposium on Intelligent and Evolutionary Systems, vol.1, pp. 653–662, 2015.

[6] W. S. Gan, “Application of neural networks to the processing ofmedical images,” in Proceedings of the IEEE International JointConference on Neural Networks (IJCNN ’91), vol. 1, pp. 300–306,IEEE, November 1991.

[7] E. S. Dunstone, “Image processing using an image approxi-mation neural network,” in Proceedings of the 1st InternationalConference on Image Processing, vol. 3, pp. 912–916, IEEE,Austin, Tex, USA, November 1994.

[8] H. S. Ranganath,G.Kuntimad, and J. L. Johnson, “Pulse coupledneural networks for image processing,” in Proceedings of theIEEE Southeastcon ’95. “Visualize the Future”, pp. 37–43, 1995.

[9] H. Tang, K. C. Tan, and Z. Yi, “Competitive neural networks forimage segmentation,” Studies in computational intelligence, vol.53, pp. 129–144, 2007.

[10] W. Xu, H. Li, and P. Xu, “A newANN-based detection algorithmof the masses in digital mammograms,” in Proceedings of theIEEE International Conference on Integration Technology (ICIT’07), pp. 26–30, IEEE, Shenzhen, China, March 2007.

[11] B. A. Alayliogh and F. Aghdasi, “An artificial neural networkfor detecting microcalcifications in wavelet-enhanced digitisedmammograms,” in Proceedings of the South African Symposiumon Communications and Signal Processing (COMSIG ’98), pp.127–132, September 1998.

[12] T. Ayer, Q. Chen, and E. S. Burnside, “Artificial neural networksin mammography interpretation and diagnostic decision mak-ing,” Computational and Mathematical Methods in Medicine,vol. 2013, Article ID 832509, 10 pages, 2013.

[13] A. P. Dhawan, Y. S. Chitre, M. Moskowitz, and G. Eric, “Clas-sification of mammographic microcalcification and structuralfeatures using an artificial neural network,” in Proceedings ofthe Annual International Conference of the IEEE Engineering inMedicine and Biology Society, vol. 13, pp. 1105–1106, 1991.

[14] B.Widrow, R. G.Winter, and R. A. Baxter, “Layered neural netsfor pattern recognition,” IEEETransactions onAcoustics, Speech,and Signal Processing, vol. 36, no. 7, pp. 1109–1118, 1988.

[15] Y. Chitre, A. P. Dhawan, and M. Moskowitz, “Artificial neuralnetwork based classification of mammographic microcalcifi-cations using image structure and cluster features,” in DigitalMammography, A.G.Gale, S. Astley, D.Dance, andA. Y. Cairns,Eds., pp. 31–40, Elsevier, 1994.

[16] S. W. Kevin, C. C. Doss, L. P. Clarke, R. A. Clark, and M. Lee,“A neural network approach to microcalcification detection,”in Proceedings of the Conference Record of the IEEE NuclearScience Symposium and Medical Imaging Conference, pp. 1273–1275, Orlando, Fla, USA, October 1992.

[17] S. Saini and R. Vijay, “Back propagation artificial neural net-work,” in Proceedings of the 5th International Conference onCommunication Systems and Network Technologies, pp. 1177–1180, Gwalior, India, April 2015.

[18] B. Zheng, Y.-H. Chang, X.-H. Wang, and W. F. Good, “Com-parison of artificial neural network and Bayesian belief networkin a computer-assisted diagnosis scheme for mammography,”in Proceedings of the International Joint Conference on NeuralNetworks (IJCNN ’99), vol. 6, pp. 4181–4185, July 1999, Cat. No.99CH36339.

[19] G. Zhang, P. Yan, H. Zhao, and X. Zhang, “A computer aideddiagnosis system in mammography using artificial neural net-works,” in Proceedings of the 1st International Conference on

BioMedical Engineering and Informatics (BMEI ’08), pp. 823–826, IEEE, Sanya, China, May 2008.

[20] A. Lashkari, “Full automatic micro calcification detection inmammogram images using artificial neural network and Gaborwavelets,” in Proceedings of the 6th Iranian Conference onMachine Vision and Image Processing (MVIP ’10), Isfahan, Iran,October 2010.

[21] D. Buller, A. Buller, P. R. Innocent, and W. Pawlak, “Determin-ing and classifying the region of interest in ultrasonic imagesof the breast using neural networks,” Artificial Intelligence inMedicine, vol. 8, no. 1, pp. 53–66, 1996.

[22] C. Ruggierol, F. Bagnolil, R. Sacilel, M. Calabrese, G. Rescinito,and F. Sardanelli, “Automatic recognition of malignant lesionsin ultrasound images by artificial neural networks,” in Proceed-ings of the 20th Annual International Conference of the IEEEEngineering inMedicine and Biology Society, vol. 2, pp. 872–875,Hong Kong, November 1998.

[23] B. Sahiner, H.-P. Chan, N. Petrick et al., “Classification of massand normal breast tissue: a convolution neural network classi-fier with spatial domain and texture images,” IEEE Transactionson Medical Imaging, vol. 15, no. 5, pp. 598–610, 1996.

[24] D.-R. Chen, R.-F. Chang, and Y.-L. Huang, “Computer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworks,” Radiology, vol. 213, no. 2, pp. 407–412, 1999.

[25] D.-R. Chen, R.-F. Chang, and Y.-L. Huang, “Breast cancer diag-nosis using self-organizing map for sonography,” Ultrasound inMedicine and Biology, vol. 26, no. 3, pp. 405–411, 2000.

[26] D.-R. Chen, R.-F. Chang, Y.-L. Huang, Y.-H. Chou, C.-M. Tiu,and P.-P. Tsai, “Texture analysis of breast tumors on sonograms,”Seminars in Ultrasound, CT andMRI, vol. 21, no. 4, pp. 308–316,2000.

[27] D.-R. Chen, W.-J. Kuo, R.-F. Chang, W. K. Moon, and C. C.Lee, “Use of the bootstrap technique with small training sets forcomputer-aided diagnosis in breast ultrasound,” Ultrasound inMedicine and Biology, vol. 28, no. 7, pp. 897–902, 2002.

[28] D.-R. Chen, R.-F. Chang, W.-J. Kuo, M.-C. Chen, and Y.-L.Huang, “Diagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networks,” Ultra-sound in Medicine and Biology, vol. 28, no. 10, pp. 1301–1310,2002.

[29] C.-M. Chen, Y.-H. Chou, K.-C. Han et al., “Breast lesionson sonograms: computer-aided diagnosis with nearly setting-independent features and artificial neural networks,” Radiology,vol. 226, no. 2, pp. 504–514, 2003.

[30] S. Joo, Y. S. Yang, W. K. Moon, and H. C. Kim, “Computer-aided diagnosis of solid breast nodules: use of an artificialneural network based on multiple sonographic features,” IEEETransactions on Medical Imaging, vol. 23, no. 10, pp. 1292–1300,2004.

[31] S. Joo, W. K. Moon, and H. C. Kim, “Computer-aided diagnosisof solid breast nodules on ultrasound with digital imageprocessing and artificial neural network,” in Proceedings of the26th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC ’04), vol. 2, pp. 1397–1400,IEEE, San Francisco, Calif, USA, September 2004.

[32] K. Zheng, T.-F. Wang, J.-L. Lin, and D.-Y. Li, “Recognition ofbreast ultrasound images using a hybridmethod,” inProceedingsof the IEEE/ICME International Conference on ComplexMedicalEngineering (CME ’07), pp. 640–643, IEEE, Beijing, China, May2007.

[33] J. Koay, C. Herry, and M. Frize, “Analysis of breast thermog-raphy with an artificial neural network,” in Proceedings of the


26th Annual International Conference of the IEEE Engineeringin Medicine and Biology Society, vol. 2, pp. 1159–1162, IEEE, SanFrancisco, Calif, USA, September 2004.

[34] T. Z. Tan, C. Quek, G. S. Ng, and E. Y. K. Ng, “A novel cognitiveinterpretation of breast cancer thermography with complemen-tary learning fuzzy neural memory structure,” Expert Systemswith Applications, vol. 33, no. 3, pp. 652–666, 2007.

[35] S. C. Fok, E. Y. K. Ng, and K. Tai, “Early detection and visuali-zation of breast tumor with thermogram and neural network,”Journal of Mechanics in Medicine and Biology, vol. 2, no. 2, pp.185–195, 2002.

[36] H. Szu, I. Kopriva, P. Hoekstra et al., “Early tumor detectionby multiple infrared unsupervised neural nets fusion,” in Pro-ceedings of the 25th Annual International Conference of the IEEEEngineering in Medicine and Biology Society, vol. 2, pp. 1133–1136, IEEE, September 2003.

[37] T. Jakubowska, B. Wiecek, M. Wysocki, C. Drews-Peszynski,andM. Strzelecki, “Classification of breast thermal images usingartificial neural networks,” in Proceedings of the 26th AnnualInternational Conference of the IEEE Engineering in MedicineandBiology Society (IEMBS ’04), vol. 2, pp. 1155–1158, September2004.

[38] A. E. Hassanien, N. El-Bendary, M. Kudelka, and V. Snasel,“Breast cancer detection and classification using support vectormachines and pulse coupled neural network,” in Proceedingsof the Third International Conference on Intelligent HumanComputer Interaction (IHCI 2011), Prague, Czech Republic,August, 2011, vol. 179 of Advances in Intelligent Systems andComputing, pp. 269–279, Springer, Berlin, Germany, 2013.

[39] G. Ertas, D. Demirgunes, and O. Erogul, “Conventional andmulti-state cellular neural networks in segmenting breast regionfrom MR images: performance comparison,” in Proceedingsof the International Symposium on Innovations in IntelligentSystems and Applications (INISTA ’12), pp. 1–4, 2012.

[40] F. A. Cardillo, A. Starita, D. Caramella, and A. Cillotti, “A neuraltool for breast cancer detection and classification in MRI,” inProceedings of the 23rd Annual International Conference of theIEEE Engineering in Medicine and Biology Society, vol. 3, pp.2733–2736, October 2001.

[41] A. A. Tzacheva, K. Najarian, and J. P. Brockway, “Breastcancer detection in gadolinium-enhanced mr images by staticregion descriptors and neural networks,” Journal of MagneticResonance Imaging, vol. 17, no. 3, pp. 337–342, 2003.

[42] A. M. ElNawasany, A. F. Ali, andM. E.Waheed, “A novel hybridperceptron neural network algorithm for classifying breastMRI tumors,” in Proceedings of the International Conference onAdvanced Machine Learning Technologies and Applications, pp.357–366, Cairo, Egypt, November 2014.

Submit your manuscripts athttps://www.hindawi.com

Stem CellsInternational

Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014


MEDIATORSINFLAMMATION

of


Behavioural Neurology

EndocrinologyInternational Journal of



Disease Markers


BioMed Research International

OncologyJournal of



Oxidative Medicine and Cellular Longevity


PPAR Research

The Scientific World JournalHindawi Publishing Corporation http://www.hindawi.com Volume 2014

Immunology ResearchHindawi Publishing Corporationhttp://www.hindawi.com Volume 2014

Journal of

ObesityJournal of



Computational and Mathematical Methods in Medicine

OphthalmologyJournal of


Diabetes ResearchJournal of



Research and TreatmentAIDS


Gastroenterology Research and Practice


Parkinson’s Disease

Evidence-Based Complementary and Alternative Medicine

Volume 2014Hindawi Publishing Corporationhttp://www.hindawi.com

Artificial Neural Networks in Image Processing for...

Documents

Transcript of Artificial Neural Networks in Image Processing for...