一种各向异性虚拟人脑图像分割模型

:167427070(2010)022011320512,,.RGB,HSL,HSV,;Gibbs,;GibbsFCM,.,.C;;;TP391141A2010201223(2006);(60973157),,,,[e-mail]1,,2100442,,2100440Introduction,,,.,.,.,,.,[1][2][3].,,[4]..(MagneticResonance,MR),3.,,,,[529].MR256256,,38722048,1,,.,,,;,,,,,,MR.,,,,RGB(Red2Green2Blue)HSL(Hue2Saturation2Lightness)HSV(Hue2Saturation2Value),.C(FuzzyC2Means,FCM),.1Fig.1Virtualhumanbrainimage1Lowerlayerdataanalyzingbasedoncolorspacein2formation,,.I,mn.RGB,W(x,y)W(x,y)=[IR(x,y),IG(x,y),IB(x,y)].RGB(R)(G)(B)3,3,RGB.:50(0,0,0)(50,0,0)2RGB,50(200,150,0)(200,200,0)2.,:CIEXYZ,CIELUV,YUV,YIQ,HSV,HSL.RGBHSLHSV:W(x,y)=014IRGB2B(x,y)+016IHSL2L(x,y)-014IHSV2V(x,y).(1)(1),,RGB2BRGBB,HSL2LHSLL,HSV2VHSVV.2,8,W..2FCMAnisotropicfuzzyclustermethod211MarkovGibbsMarkov,XN={Ni,iX}.Nii,Nii;2Fig.2ImageconversioniNj,jNi.F,P(x)0,PxF,PxixX-i=PxixNi,XMarkov.MRF,.Hamersley2Clifford,FK,MRFGibbs:P(x)=exp(-U(x))xFexp(-U(x)).(2),U(x),U(x)=cKVc(x),Vc(x)xc,V(xi-xj)=1-(xi-xj),,dirac.Gibbs,:Ni=xjxj-xiR,ji,R.,Gibbs,.,,Gibbs.J=I2xIxIyIxIyI2y.(3)J,,1=12J11+J22+(J11-J22)2+4J212;2=12J11+J22-(J11-J22)2+4J212.1=[cossin]T;2=[-sincos]T.411,..WANGShunfeng,etal.Ananisotropicvirtualhumanbrainimagesegmentationmodel.cos=(2J12)/Jnor;sin=J22-J11+(J11-J22)2+4J212Jnor;Jnor=(2J12)2+J22-J11+(J11-J22)2+4J2122.,,.,,,.4J=G3J11G3J12G3J21G3J22.(4):G,3.,JJ.,,:5tJij=divg2k,l¨Jk,l¨JTk,l¨Ji,j;g(x)=1/(1+x2)i,j=1,2.(5)J,,.,1,2,1,2,0,,1=;2=,1=2;(1-exp(-1/(1-2)2)).Ni=xjxjj,ji,,jxi.xi(ai,bi),xj(aj,bj),((aj-ai)cos+(bj-bi)sin)2(1)2+(-(aj-ai)sin+(bj-bi)cos)2(2)21,(6)xjNi.,,;,;,.:,120;1m20;120.,,Gibbs,,Gibbs.212FCM,.FCM,,,T=i,jKk=1uk(i,j)2I(i,j)-vk2(7),.(7):I(i,j)(i,j);vkk,k;uk(i,j)(i,j)k;K;.I=IB,,I,I,B,.loglog(I)=log(IB)=log(I)+log(B).(8),log(I)I,log(I)I,log(B)B,I=I+B.xiNi,xiP(xi),Ni1-P(xi).,(7)T=i,jKk=1(1-Pk(i,j))uk(i,j)2I(i,j)-B(i,j)-vk2.(9),Pk(i,j)N(i,j),(i,j)k,,I(i,j)-vk,(1-Pk(i,j)),.,(9)uk(i,j)=I(i,j)-B(i,j)-vk-2(1-Pk(i,j))-2Kl=1I(i,j)-B(i,j)-vl-2(1-Pl(i,j))-2;(10)vk=i,juk(i,j)2(1-Pk(i,j))(I(i,j)-B(i,j))i,juk(i,j)2(1-Pk(i,j));(11)B(i,j)=Ck=1uk(i,j)2(1-Pk(i,j))(I(i,j)-vk)Ck=1uk(i,j)2(1-Pk(i,j)).(12),Gibbs,511:,2010,2(2):1132117JournalofNanjingUniversityofInformationScienceandTechnology:NaturalScienceEdition,2010,2(2):1132117,.S=det(J)tr(J)=121+2.(13):S()mS()S()0.,1,,.3Resultsandanalysis3a,256256,,FCM,3bFCM3,FCM,;3c,,,,,.3Fig.3Classificationresultsofvirtualimage4112371445.4a,,,.BrainSurfaceExtractor(BSE)[10].4b.,,BSE.,,RGB.4c,,.5,5aBrainSurfaceExtractor,,,.5b4Fig.4Imagesofshelledvirtualhumanbrain.,,GibbsFCM,.5c5d.5Fig.5Resultsofthree2dimensionshelling611,..WANGShunfeng,etal.Ananisotropicvirtualhumanbrainimagesegmentationmodel.:Q(S1,S2)=|S1S2||S1S2|.(16),S1,S2,1.1Table1Segmentationprecision/%/%BrainSurfaceExtractor761696313599124981964ConclutionRGB,HSL,HSV,.Gibbs,GibbsFCM,,.References[1]BoesenK,RehmK,SchaperK,etal.Quantitativecomparisonoffourbrainextractionalgorithms[J].Neuroimage,2004,22(3):125521261[2]LeGG,ArgentiAM,DuymeM,etal.Statisticalsulcalshapecomparisons:Applicationtothedetectionofgeneticencodingofthecentralsulcusshape[J].Neuroimage,2000,19(11):5642574[3]LogothetisNK,PaulsJ,AugathM,etal.Neurophsiologicalinves2tigationofthebasisoftheFMRIsignal[J].Nature,2001,412(12):1502157[4]WoermannFG,SisodiyaSM,FreeSL,etal.QuantitativeMRIinpatientswithidiopathicgeneralizedepilepsy:Evidenceofwide2spreadcerebralstructuralchanges[J].Brain,1998,121(5):166121667[5]KinkinsR,ShentonME,GerigG,etal.Routinequantitativeanal2ysisofbrainandcerebrospinalfluidvolumeswithMRimaging[J].MagneticResonanceImaging,1992,2(6):619229[6]GerigG,MartinJ,KikinisR,etal.Unsupervisedtissuetypeseg2mentationof3Ddual2echoMRheaddata[J].ImageVisionCom2pute,1992(10):3492360[7]AtkinsMS,MackiewichBT.FullyautomaticsegmentationofthebraininMRI[J].IEEETransactionsonMedicalImaging,1998,17(1):982107[8]DaleAM,FischlB,SerenoMI.Corticalsurface2basedanalysis.I.Segmentationandsurfacereconstruction[J].Neuroimage,1999,9(2):1792194[9]StokkingR,VinckenKL,ViergeverMA.Automaticmorphology2basedbrainsegmentation(MBRASE)fromMRI2T1data[J].Neu2roimage,2000,12(6):7262738[10]CardenasaVA,EzekielF,SclafanidVD,etal.Reliabilityoftis2suevolumesandtheirspatialdistributionforsegmentedmagneticresonanceimages[J].PsychiatryRes:Neuroimage,2001,106(3):1932205AnanisotropicvirtualhumanbrainimagesegmentationmodelWANGShunfeng1LUOChunyan21CollegeofBinjiang,NanjingUniversityofInformationScience&Technology,Nanjing2100442SchoolofMath&Physics,NanjingUniversityofInformationScience&Technology,Nanjing210044AbstractVirtualbraintissueinformationextractionhasbecomeanimportantpartofvirtualhumanbraindataa2nalysis.However,traditionalextractionmethodscannotobtainsatisfactoryresultsforimagenoiseandlowerlayerda2tadistraction.Inthispaper,RGB,HSLandHSVspaceinformationwereusedtoconstructanewinformationfield,whichcanreducetheimpactoflowerlayerdata.ThenananisotropicGibbsfieldwasbuiltwithstructuretensorin2formationtoreducetheeffectofnoise.TheimprovedFCMmodelwithanisotropicGibbsfieldwasintroducedtoseg2mentimage,inordertominimizetheerrorcausedbyintensityinhomogeneity.Experimentresultsindic