资源描述：
《A study on several machine-learning methods》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。
1、IEEETRANSACTIONSONMEDICALIMAGING,VOL.24,NO.3,MARCH2005371AStudyonSeveralMachine-LearningMethodsforClassificationofMalignantandBenignClusteredMicrocalcificationsLiyangWei,StudentMember,IEEE,YongyiYang*,SeniorMember,IEEE,RobertM.Nishikawa,andYuleiJiangAbstract—Inthispaper,weinvestigateseve
2、ralstate-of-the-artmammograms.ClusteredMCscanbeanimportantearlyindi-machine-learningmethodsforautomatedclassificationofclus-catorofbreastcancerinwomen.Theyappearin30%–50%ofteredmicrocalcifications(MCs).Theclassifierispartofacom-mammographicallydiagnosedcases.Forexample,Fig.1showsputer-aid
3、eddiagnosis(CADx)schemethatisaimedtoassistingamammogramwithaclusterofMCs.Thoughcommonlyseenradiologistsinmakingmoreaccuratediagnosesofbreastcanceronmammograms.Themethodsweconsideredwere:supportonmammograms,MCsareoftendifficulttodiagnoseaccuately.vectormachine(SVM),kernelFisherdiscrimina
4、nt(KFD),rele-Thisgreatlycompromisesthequalityofradiologists’biopsyvancevectormachine(RVM),andcommitteemachines(ensemblerecommendations,whichisanimportantissueinbreastcanceraveragingandAdaBoost),ofwhichmosthavebeendevelopedre-diagnosis.Itisreportedthatamongthosewithradiographicallycentl
5、yinstatisticallearningtheory.Weformulateddifferentiationsuspicious,nonpalpablelesionswhoaresentforbiopsy,onlyofmalignantfrombenignMCsasasupervisedlearningproblem,15%–34%arefoundtoactuallyhavemalignancies[1],[2].andappliedtheselearningmethodstodeveloptheclassificationalgorithm.Asinput,th
6、esemethodsusedimagefeaturesautomat-Therehasbeenagreatdealofresearchinrecentyearstode-icallyextractedfromclusteredMCs.Wetestedthesemethodsvelopcomputerizedmethodsthatpotentiallycouldassistradi-usingadatabaseof697clinicalmammogramsfrom386cases,ologistsindifferentiatingbenignfrommalignant
7、MCs.Usingwhichincludedawidespectrumofdifficult-to-classifycases.Weacomputer-aideddiagnosis(CADx)scheme,radiologistscouldanalyzedthedistributionofthecasesinthisdatabaseusingtheincorporatetheoutputfromthecomputerintotheirdecision.multidimensionalscalingtechnique,whichrevealsthatinthefea
