基于Graph Cuts算法的乳腺X線圖像腫塊分割方法研究.pdf

1、杭州電子科技大學碩士學位論文基于GraphCuts算法的乳腺X線圖像腫塊分割方法研究姓名：吳相穎申請學位級別：碩士專業(yè)：模式識別與智能系統(tǒng)指導教師：厲力華；徐偉棟201112ABSTRACTBreastcaIlcerremaillsam句0rcauseofc鋤cerdeathsamongwomeIl鯽孤nd也eworldMammo掣印hyiscllrreIltlyacc印tedasthemoste行ectiVemeasuref醯eall

2、ydetection觚ddia印oslsofbreastc鋤ceLIIlordertoimproVee伍ci％cy鋤daocuracyinmalT姍。黟aphicbreastcallcerdetectioⅣdia鯛osis，theComl)utaideddiagnosis(CAD)tecIlIlologyhasbeeIlwidelyllsedbyradiolo西stsAsa11importantstepinbreastXrayimagc

3、CADsyst鋤，se舯咖atlonreSultotthemassreflectspath010西cfeatIlresofthemassanddirectlya仃eCtssubsequentfeatureex仃actlonandclassificationHoweV％nlerearem鋤ynoise鋤danifactintllem撒Ⅱno薩鋤，and也erelslowcontr勰tb咖eencomplexmassandmebackgro

4、undExistingmethodsi芏lmasssegm饑tatlonareoReIlinsu伍cienttomeetdinicalrequir鋤entsGrapbC懈h舔beeIluSedinrec饑tyearsforinteractiVeimagesegrn饑tatlon。l。hccoreideologyofGra】phCutsistom印趾imageontoanet、)Irork黟印h，aIldConstnlctaIlelle暇

5、，劬ctlon徹melabelin蜀a11dmelldo鋤朗嘿：)rminimizationwimcombinatorialopt砌zationtecl瓶quesTllisstudyproposesanewse舯耐ationm咖0duSing豫atcdGraphC幽baSedonmulti’scale鋤oomingne刪ltiscalem砒odcallse舯饑tm鋤mo黟印hicimageswlthast印wlseprocess舶m西o

6、baltolocalsegmentationbyiteratingGr印hCutsAnu11supeⅣisedwat蹦med仃ansfo冊ofthemorpholo西cal黟adientofthe麗西nalXrayimageisusedinourworkt0produceare晷onadiaceIlcy擎aphforme叩timizationst印sMoreoVerilIlponantcomou】噶ot也emaIllmo腫pmcimag

7、earepreseⅣeddumgthe6rStunSupeⅣisedse舯饑tation111enewse舯eIltatlons蹦eg)，e舵ctivelyimprovesse舯entationpe南mallcewimlessiIlnua∞esofimagcnoiselevel110improveinsu衢cieIlciesofthemultiscalememod，suchas西obalmiIlimizationoft11eeIlerg

8、y缸lctionsisNPhardeveIlintllesimplestdiscominui妒preservingcase；thistllesis咖dycontinuestoin咖ducea觚，automaticand雒試entsegm饑tationm以lodFuzzyCmeaIlsClust耐ngalgoritllIIlisusedforpre—se舯eIltationofmemasstoproVideaIliIlitialmadkf

9、ieldfortllealgorithmaIldthell口一expaIlsionalgorithmbased0nGra【phCutsquicklyse叭Iheslocalminlmumofmeener斟鼬lction，a11dconductS蠡撾approXilIlate鋤哪黟0ptinAization111eproposedmettmimprovescalculatione銜ciellcy黟eatlywhileke印ingsegme

10、Iltationpreclslon。I。heeXpenmentaIresultsd鋤onstratethattlleproposedmethodcanproVideaccurate觚dhi曲lyrobustsegmentatlonresultsformammo鏟aphicimagewithinteIlsityiIIhomogeneityandhe剁rynoiseKeywords：M鋤mogram，MassSe腫entation，Grap