基于人工神经网络对肾结石病的诊断(IJITCS-V4-N7-3)

I.J.InformationTechnologyandComputerScience,2012,7,20-25PublishedOnlineJuly2012inMECS()DOI:10.5815/ijitcs.2012.07.03Copyright©2012MECSI.J.InformationTechnologyandComputerScience,2012,7,20-25ArtificialNeuralNetworksforDiagnosisofKidneyStonesDiseaseKoushalKumarDepartmentofCSE/IT,LovelyProfessionalUniversity,Jalandhar,Punjab,[e-mail]SE/IT,LovelyProfessionalUniversity,Jalandhar,Punjab,[e-mail]—ArtificialNeuralnetworksareoftenusedasapowerfuldiscriminatingclassifierfortasksinmedicaldiagnosisforearlydetectionofdiseases.Theyhaveseveraladvantagesoverparametricclassifierssuchasdiscriminateanalysis.Theobjectiveofthispaperistodiagnosekidneystonediseasebyusingthreedifferentneuralnetworkalgorithmswhichhavedifferentarchitectureandcharacteristics.Theaimofthisworkistocomparetheperformanceofallthreeneuralnetworksonthebasisofitsaccuracy,timetakentobuildmodel,andtrainingdatasetsize.WewilluseLearningvectorquantization(LVQ),twolayersfeedforwardperceptrontrainedwithbackpropagationtrainingalgorithmandRadialbasisfunction(RBF)networksfordiagnosisofkidneystonedisease.InthisworkweusedWaikatoEnvironmentforKnowledgeAnalysis(WEKA)version3.7.5assimulationtoolwhichisanopensourcetool.Thedatasetweusedfordiagnosisisrealworlddatawith1000instancesand8attributes.Intheendpartwechecktheperformancecomparisonofdifferentalgorithmstoproposethebestalgorithmforkidneystonediagnosis.Sothiswillhelpsinearlyidentificationofkidneystoneinpatientsandreducesthediagnosistime.IndexTerms—KidneyStoneDisease,MultilayerPerceptrons,RadialBasisFunctionNetworks,LearningVectorQuantization,DiagnosisI.IntroductionKidneystonesdiseaseisbecomingmoreandmorecommonnowdays.Kidneystonesarecreatedwhencertainsubstancesinurineincludingcalcium,oxalate,andsometimesuricacidcrystallize.Thesemineralsandsaltsformcrystals,whichcanthenjointogetherandformakidneystone.Eachtypeofkidneystonehasadifferentcause[1].Stonesareclassifiedaccordingtotheirchemicalcomposition.Approximately80%ofallkidneysstonesarecalciumoxalatestones,whicharethemostproblematic.Theformationofthesestonesmaybecausedbygeneticfactorsandalsodependsuponage,andgeographicalfactors[2].However,moreimportantaredietaryandlifestylefactors,andtheresultsofacquiredmetabolicdefectsleadingtocrystalformationandgrowthofakidneystone[3].In[2]authorsgiveadetailexplanationregardingwhatarekidneystones,itsTypesanddifferentsymptomsofthisdisease.In[3]authorsdescribethedifferentfactorslikeage,sex,race,bodyweight,ethnicitywhichmaycauseofkidneystones.ThemostcommonproblemintheFieldofautomaticdiagnosticisthediagnosticsusingfastandaccuratealgorithmwhichdoesn’trequirelongtimetorunandgiveaccurateandcorrectresults[4].Toreducethediagnosistimeandimprovethediagnosisaccuracy,ithasbecomemoreofademandingissuetodevelopreliableandpowerfulmedicaldiagnosissystemtosupporttheyetandstillincreasinglycomplicateddiagnosisdecisionprocess.ThemedicaldiagnosisbynatureisacomplexandfuzzycognitiveProcesshencesoftcomputingmethods,suchasneuralnetworks,haveshowngreatpotentialtobeappliedinthedevelopmentofmedicaldiagnosis.Indiseasediagnosisthelearninganddetectionofpartialdiseasecanbehelpfulwhentimeandinformationconstraintsarepresent.Thusartificialneuralnetworksprovideagoodmeanstopartialdiagnosis.Thisweusedthreeneuralnetworksalgorithmsformeasuringtheirclassificationaccuracyagainsttimetakentoclassifyfordiagnosispurpose.ThispaperisthusorganizedasfollowinginsectionIIabriefintroductionoftheartificialneuralnetwork,insectionIIIPreviousrelatedworkworksthathadbeendone,insectionIVkidneystonedatasetthatisusedinthisresearchhasbeendiscussed,insectionVartificialneuralnetworksclassifiersusedisdescribed,InsectionVIsimulationtoolusedisdescribed,insectionVIIexperimentresultsanddiscussionisgivenandinthelastsectionweconcludethepaper.ArtificialNeuralNetworksforDiagnosisofKidneyStonesDisease21Copyright©2012MECSI.J.InformationTechnologyandComputerScience,2012,7,20-25II.ArtificialNeuralNetworksIntroductionArtificialneuralnetworks(ANN)haveemergedasaresultofsimulationofbiologicalnervoussystem,suchasthebrainonacomputer.ArtificialNeuralnetworksarerepresentedasasetofnodescalledneuronsandconnectionsbetweenthem.Theconnectionshaveweightsassociatedwiththem,representingthe―strength‖ofthoseconnections.Nowadaysneuralnetworkscanbeappliedtoproblemsthatdonothavealgorithmicsolutionsorproblemsforwhichalgorithmicsolutionsaretoocomplextobefound.Inotherswordsthekindofproblemsinwhichinputsandoutputsvariablesdoesnothaveaclearrelationshipbetweenthem,aneuralnetworksisaefficientapproachinsuchproblems.Mostneuralnetworkarchitecturehasthreelayersinitsstructure.Firstlayerisinputlayerwhichprovidesaninterfacewiththeenvironment,secondlayerishiddenlayerwherecomputationisdoneandlastlayerisoutputlayerwhereoutputisstored.Dataispropagatedthroughsuccessivelayers,withthefinalresultavailableatthe―outputlayer‖.Manydifferenttypesofneuralnetworksareavailableandmultilayerneuralnetworksarethemostpopular.MLPpopularityisduetomorethenonehiddenlayerinitsstru