肝硬化數(shù)據(jù)挖掘

ComparisonofAITechniquesforPredictionofLiverFibrosis

inHepatitisPatientsJournalofMedicalSystemJiajunShiSomeexplanationsFibrosis-纖維化Hepatitis-肝炎HepatitisB/C–乙肝/丙肝Cirrhosis–肝硬化Liverbiopsies-活組織檢查Non-invasivetechniques–無創(chuàng)技術Serummarkers–血清標記

OutlineIntroductionBackground:AIandCDSSNa?veBayesClassifier(NBC)&LogisticsRegressionHepatitisandFibrosisStageAIAssistedWeb-basedClinicalDecisionSupportSystemFourMethodsResultsandDiagnosticAccuracyConclusionIntroductionOneintwelvepeoplehavetheHepatitisBorHepatitisCvirusDiagnosisandtreatmentofthisdiseaseisguidedbyliverbiopsieswhereasmallamountoftissueisremovedbyasurgeonandexaminedbyapathologistDeterminethefibrosisstagefromF0(nodamage)toF4(cirrhosis)RiskandcostlyNon-invasivetechniques,withserummarkers,imagingtest,andgeneticstudiesAccuracynotachievedsufficientacceptanceIntroductionNon-invasivetechniques,withserummarkers,imagingtest,andgeneticstudiesAI

&CDSSKnowledgeofthelevelofliverdamageinapatientwith

liverdisease(particularlyHepatitisBandHepatitisC)isa

criticalfactorindeterminingtheoptimalcourseoftreatment

andtomeasuretheeffectivenessofalternativetreatmentsin

patients.NotaccurateBackgroundofAIandCDSSArtificialIntelligenceandDataMiningtechniquesIncludeNeuralNetworks,FuzzyLogic,DecisionTrees,BayesianClassifiers,SupportVectorMachines,GeneticAlgorithmsandHybridSystemClinicalandMedicalDecisionSupportSystemsSupporttheprocessofdiscoveringusefulinformationinlargeclinicalrepositoriesTheyhaddonethesystemdesignedwithneuralnetworksanddecisiontreemethodsbecauseoftheirsuccessfulapplicationinsimilarproblemdomainsHepatitisandFibrosisStageOneintwelvepeoplehavetheHepatitisBorHepatitisCvirusFibrosisStage

Description0Nofibrosis-Normalconnectivetissue

1Portalfibrosis-Fibrousportalexpansion

2Periportalfibrosis-Periportalorrareportal-portalsepta

3Septalfibrosis-Fibrousseptawitharchitecturaldistortion;no

obviouscirrhosis

4Cirrhosis

AIAssistedWeb-basedClinicalDecisionSupportSystemAIAssistedCDSSAItechniquesResultingknowledgebaseAIAssistedWeb-basedClinicalDecisionSupportSystemExplanations血清細胞堿性磷酸酶血清膽堿酯酶膽紅素谷氨酸轉肽酶丙種球蛋白類測試時年齡乙肝or丙肝Variables：SerumMarkersPatientsInfoAIAssistedweb-basedClinicalDecisionSupportSystemSysteminputs&Outputs:FourMethodsPaper‘AdvancedDecisionSupportforComplexClinicalDecisions’NeuralNetworks,DecisionTreesThispaperNaiveBayesandLogRegressionMethodinputs:FourMethods–Na?vebayesclassifierThevariationinmeanvaluesfortwoparameters(ABLandG-GL)areshownbyfibrosisstageintheFigure.Withthismodel,wecancalculatethecombinedprobabilityofeachfibrosisstagethenselectthehighestprobableasourpredictedresult.FourMethods-LogisticsregressionCrossValidationandDiagnosticAccuracyCrossValidationandDiagnosticAccuracyAccuracyofFibrosisStagePredictions(424patients)

PredictiveSensitivityandSpecificityConclusionThefourartificialintelligencemethodspresentedinthisstudyshowedsomesignificantvariabilityinaccuracy,sensitivity,andspecificityinpredictingfibrosisstageindataon424hepatitispatients.Althoughneuralnetworkmethodsshowedthehighestsensitivityandspecificity,theirroleispredictingtheexactfibrosisstagewasrelativelypoor.Logisticregressionandna?vebayesmethodswereth