基于BP神经网络的脑肿瘤的分类(IJIGSP-V5-N2-7)

I.J.Image,GraphicsandSignalProcessing,2013,2,45-50PublishedOnlineFebruary2013inMECS()DOI:10.5815/ijigsp.2013.02.07Copyright©2013MECSI.J.Image,GraphicsandSignalProcessing,2013,2,45-50BrainTumorClassificationUsingBackPropagationNeuralNetworkN.SumitraDepartmentofInstrumentationandControlEngineering,GautamBuddhTechnicalUniversityJSSAcademyofTechnicalEducation,Noida(U.P),IndiaE-mail:[e-mail]rtmentofElectronicInstrumentationandControlEngineering,RajasthanTechnicalUniversityInstituteofEngineeringandTechnology,Alwar(Raj.),IndiaE-mail:[e-mail]—Theconventionalmethodformedicalresonancebrainimagesclassificationandtumorsdetectionisbyhumaninspection.Operator-assistedclassificationmethodsareimpracticalforlargeamountsofdataandarealsonon-reproducible.Hence,thispaperpresentsNeuralNetworktechniquesfortheclassificationofthemagneticresonancehumanbrainimages.TheproposedNeuralNetworktechniqueconsistsofthefollowingstagesnamely,featureextraction,dimensionalityreduction,andclassification.Thefeaturesextractedfromthemagneticresonanceimages(MRI)havebeenreducedusingprinciplescomponentanalysis(PCA)tothemoreessentialfeaturessuchasmean,median,variance,correlation,valuesofmaximumandminimumintensity.Intheclassificationstage,classifierbasedonBack-PropagationNeuralNetworkhasbeendeveloped.Thisclassifierhasbeenusedtoclassifysubjectsasnormal,benignandmalignantbraintumorimages.TheresultsshowthatBPNclassifiergivesfastandaccurateclassificationthantheotherneuralnetworksandcanbeeffectivelyusedforclassifyingbraintumorwithhighlevelofaccuracy.IndexTerms—Backpropagationneuralnetwork,PCA,Malignant,BenignI.INTRODUCTIONEarlydetectionandclassificationofbraintumorsisveryimportantinclinicalpractice.Manyresearchershaveproposeddifferenttechniquesfortheclassificationofbraintumorsbasedondifferentsourcesofinformation.Inthispaperweproposeaprocessforbraintumorclassification,focusingontheanalysisofMagneticResonance(MR)imagesdatacollectedwithbenignandmalignanttumors.Ouraimistoachieveahighaccuracyindiscriminatingthetwotypesoftumorsthroughacombinationofseveraltechniquessuchasimageprocessing,featureextractionandclassification.Theproposedtechniquehasthepotentialofassistingclinicaldiagnosis.Imageprocessingtechniquesareplayingimportantroleinanalysinganatomicalstructuresofhumanbody.Imageacquisitiontechniqueslikemagneticresonanceimaging(MRI),X-Ray,ultrasound,mammography,CT-scanarehighlydependentoncomputertechnologytogeneratedigitalimages.Afterobtainingdigitalimages,imageprocessingtechniquescanbefurtherusedforanalysisofregionofinterest.Atumorisamassoftissuethatservesnousefulpurposeandgenerallyexistsattheexpenseofhealthytissues.Malignantbraintumorsdonothavedistinctborders.Theytendtogrowrapidly,increasingpressurewithinthebrainandcanspreadinthebrainorspinalcordbeyondthepointwheretheyoriginate.Theygrowfasterthanbenigntumorsandaremorelikelytocausehealthproblems.Benignbraintumors,composedofharmlesscells,haveclearlydefinedborders,canusuallybecompletelyremovedandareunlikelytorecur.Abenigntumorisbasicallyatumorthatdoesn'tcomebackanddoesn'tspreadtootherpartsofthebody.Benigntumorstendtogrowmoreslowlythanmalignanttumorsandarelesslikelytocausehealthproblems.InbrainMRimages,afterappropriatesegmentationofbraintumorclassificationoftumorintomalignantandbenignisdifficulttaskduetocomplexityandvariationsintumortissuecharacteristicslikeitsshape,size,graylevelintensitiesandlocation[1].Featureextractionisanimportantissueforanypatternrecognitionproblem.Mostofthereportedworkisdedicatedtotumorsegmentationortumordetection[1-6].Thispaperpresentsahybridapproachtoclassifymalignantandbenigntumorsusingsomepriorknowledgelikepixelintensityandsomeanatomicalfeaturesareproposed.MATLAB®7.01,itsimage46BrainTumorClassificationUsingBackPropagationNeuralNetworkCopyright©2013MECSI.J.Image,GraphicsandSignalProcessing,2013,2,45-50processingtoolboxisusedforfeatureextractionandANNtoolboxhasbeenusedforclassification.Theoverallorganizationofthepaperisasfollows.ThestepsusedtoextracttheprincipalfeaturesusingPrincipalcomponentanalysisareexplainedinSectionII.SectionIIIdescribestheproposedmethodologyusedandineachsubsectionithasbeenexplainedindetails.SectionIVdemonstratessomesimulationresultsandtheirperformanceevaluationfinallyconclusionsarepresentedinsectionVIwhichtellstheadvantagesofthiswork.Someotherfutureworksinthisfieldhasbeenproposedinthisparttoo.II.FEATUREEXTRACTIONUSINGPRINCIPALCOMPONENTANALYSIS(PCA)Inthispaper,theprincipalcomponentanalysis(PCA)isusedasafeatureextractionalgorithm.Theprincipalcomponentanalysis(PCA)isoneofthemostsuccessfultechniquesthathavebeenusedinimagerecognitionandcompression.ThepurposeofPCAistoreducethelargedimensionalityofthedata.Insteadofemployingallofthefeatures,apreprocessingstepoffeatureselectionisperformedusingPCAaimingtoremovetheredundantfeatures.Basedonthestatisticalinformation,onlythemostinformativefeaturesextractedfromtheMRimagesareutilizedintheprocess.Thestepsusedtoextracttheprincipalcomponentsusingprinci