基于卷积神经网络的皮肤病网络诊断(IJITCS-V11-N11-6)

I.J.InformationTechnologyandComputerScience,2019,11,54-60PublishedOnlineNovember2019inMECS()DOI:10.5815/ijitcs.2019.11.06Copyright©2019MECSI.J.InformationTechnologyandComputerScience,2019,11,54-60AWeb-BasedSkinDiseaseDiagnosisUsingConvolutionalNeuralNetworksSamuelAkyeramfo-SamDepartmentofComputerScience,SunyaniTechnicalUniversity,SunyaniGhanaE-mail:[e-mail]epartmentofComputerScience,SunyaniTechnicalUniversity,SunyaniGhanaE-mail:[e-mail]ntofComputerScience,SunyaniTechnicalUniversity,SunyaniGhanaE-mail:[e-mail]epartmentofComputerScience,SunyaniTechnicalUniversity,SunyaniGhanaE-mail:[e-mail]tofComputerScience,SunyaniTechnicalUniversity,SunyaniGhanaE-mail:[e-mail]:07June2019;Accepted:28September2019;Published:08November2019Abstract—Skindiseasesarereportedtobethemostcommondiseaseinhumansamongallagegroupsandasignificantrootofinfectioninsub-SaharanAfrica.Thediagnosisofskindiseasesusingconventionalapproachesinvolvesseveraltests.Duetothis,thediagnosisprocessisseentobeintenselylaborious,time-consumingandrequiresanextensiveunderstandingofthedomain.Theenhancementofcomputervisionthroughartificialintelligencehasledtoamorestraightforwardandquickerwayofdetectingpatternsinimages,whichcanbeharnessedtoequipdiagnosisprocess.Despitethebreakthroughintechnology,thedermatologicalprocessinGhanaisyettobeautomated,makingthediagnosisprocesscomplicatedandtime-consuming.Hence,thisstudysoughttoproposeaweb-basedskindiseasedetectionsystemnamedmedilab-plususingaconvolutionalneuralnetworkclassifierbuiltupontheTensorflowframeworkfordetecting(atopicdermatitis,acnevulgaris,andscabies)skindiseases.Experimentalresultsoftheproposedsystemexhibitedclassificationaccuracyof88%foratopicdermatitis,85%foracnevulgaris,and84.7%forscabies.Again,thecomputationaltime(0.0001seconds)oftheproposedsystemimpliesthatanydermatologist,whodecidestoimplementthisstudy,canattendtonotlessthan1,440patientsadaycomparedtothemanualdiagnosisprocess.Itisestimatedthattheproposedsystemwillenhanceaccuracyandofferfastingdiagnosisresultsthanthetraditionalmethod,whichmakesthissystematrustworthyandresourcefulfordermatologicaldiseasedetection.Additionally,thesystemcanserveasarealtimelearningplatformforstudentsstudyingdermatologyinmedicalschoolsinGhana.IndexTerms—Skindiseasedetection,Expert-system,Convolutionalneuralnetwork,Tensorflow,Atopicdermatitis,Acnevulgaris,Scabies.I.INTRODUCTIONThehumanskinplaysahugepartinaperson'sphysicalappearance,anditisthebiggestorganofthehumanbody.Thehuman-skinoffersprotectionagainstfungalinfection,bacteria,allergy,virusesandcontrolsthetemperatureofthebodysituationsthatchangethetextureoftheskinordamagetheskincanproducesymptomslikeswelling,burning,redness,anditching[1].Antipathies,geneticstructure,irritants,andparticulardiseasesandimmunesystemassociatedcomplicationscanproducehives,dermatitis,andotherskinproblems.Manyoftheskindiseases,suchasalopecia,eczema,acne,ringworm,alsoaffectaperson’slook.Skindiseasesarewidespreadthesedays;someofthemaresimpleandeasytorecoverfrom;othersareveryharmfulandmightbeincurable,andmanyofthesediseasesareverydangerous,mainlyifnottreatedintheearlystages.Inliterature,dermatologicaldiseasesarereportedtobethemostwidelyspreaddisease[2].AsurveybyHogewoningetal.[3]revealedthatthetotalfrequencyofAWeb-BasedSkinDiseaseDiagnosisUsingConvolutionalNeuralNetworks55Copyright©2019MECSI.J.InformationTechnologyandComputerScience,2019,11,54-60pupilswithsomeskindiseasewas34.6%and42.0%intwo(2)Ghanaianstudies,outof4,839pupilssurveyed.Again,in[4]reports,outof529participantssurveyed,700discreteskindiagnosesweremade[4].Therefore,detectionofskindiseaseatitsearlystagesisparamounttoitsspreading.Ontheotherhand,skindiseasediagnosisisseentobecomplicated,mainlywhentwoormorediseasesportraysameorsimilarsymptoms,hencerequiresadermatologistwithvastexperienceofskindiseases[2,4].Nevertheless,thedevelopmentintechnologyandmachinelearninghavechangedallaspectsofone’sday-to-daylife,includingthemedicalfield[5,6].Manytherapeuticsystemshavebeendevelopedwiththehelpofartificialintelligence(AI)andtechnologicaladvancementtohelpbothdoctorsandpatientsindiverseways,startingfromOutPatientDepartment(OPD),consultationtotheoperatingtheatreoroperatingroom(OR).Thus,theintroductionofartificialintelligenceintothehealthindustrieshasbroughttremendousimprovementinthediagnosesofskindiseaseandotherillness[7].However,inGhana,mostdermatologistsstilluseavarietyofmanualvisualcluessuchascolour,scaling,andarrangementofthelesions,thebodysitedistribution,amongothers.Nonetheless,whentheseindividualcomponentsareanalysedseparately,therecognitionofthediseasecanbequitecomplex,thusrequiringahighlevelofexperience.Humandiagnosisisbasedonasubjectivejudgmentofthedermatologist,soitishardlyreproducible,unlikecomputer-aideddiagnosticsystems,whicharemorerealisticandreliable.Toreducediagnosistimeandprovidequickhealthservice,someresearcher