纹理分析在生物组织光学相干层析图像信息提取和特征识别中的应用

48,011701(2011)Laser&OptoelectronicsProgress2011纹理分析在生物组织光学相干层析图像信息提取和特征识别中的应用梁艳梅张舒(,300071)(OCT),OCT,,,,,;;;;;TN911.73Adoi:10.3788/LOP48.011701TextureAnalysisforInformationExtractionandFeatureRecognitioninOpticalCoherenceTomographyImagesLiangYanmeiZhangShu(KeyLaboratoryofOptoElectronicInformationScienceandTechnology,EducationMinistryofChina,InstituteofModernOptics,NankaiUniversity,Tianjin300071,China)AbstractWiththedevelopmentoftheopticalcoherencetomography(OCT)inthefieldofbiomedicalimaging,computeraidedmedicaldiagnosisandtreatmenteffectivenessevaluationbymeansoftherelevanttissuefeaturesreflectedintheOCTimage,hasattractedmuchattention.AmongmethodsaimedattheinformationextractionandfeaturerecognitioninOCTimages,textureanalysishasalreadybeencoveredthoroughlyandshowedagoodfeasibility.Inthispaper,weconcentrateonthecharacteristicsandapplicationsofvarioustextureanalysismethods,followedbytheexistingproblemsandpossiblesolutions.Keywordsbiooptics;opticalcoherencetomography;informationextractionandfeaturerecognition;textureanalysis;cooccurrencematrices;machinerecognitionOCIScodes170.4500;100.2960;170.6935;100.4966:20100407;:20100830:(60677012,60637020)(09JCZDJC18300):(1970-),,,Email:[e-mail]1(OCT)[1],,OCTX,,,[2],OCT,,,1996OCT,,OCTOCT011701148,011701[3],,OCT,[4]OCT,,OCT,,,OCT,2,,,,[5],,(MRI)(CT),[6][7][8],[912],[13,14],,[15,16],,,,2003,Gossage[17]OCT,OCTOCTs(i,j|d),i,(=0,/4,/2,3/4),jd,E1,E2,CHI,E1=!L-1i=0!L-1j=0[s(i,j|d)]2,(1)E2=!L-1i=0!L-1j=0s(i,j|d)lg[s(i,j|d)],(2)C=!L-1i=0!L-1j=0[(i-x)(j-y)s(i,j|d)]/(!x!y),(3)H=!L-1i=0!L-1j=011+(i-j)2s(i,j|d),(4)I=!L-1i=0!L-1j=0(i-j)2s(i,j|d),(5),L,x=!L-1i=0i!L-1j=0s(i,j|d),(6)y=!L-1i=0j!L-1j=0s(i,j|d),(7)!x=!L-1i=0(i-x)2!L-1j=0s(i,j|d),(8)!y=!L-1i=0(j-y)2!L-1j=0s(i,j|d).(9),,24,3X,X=(X~-~)/!~,(10)011701248,011701~,~!~,cdc=(X-c)T!c(X-c),(11),!c[m][n]=(X[m]-c[m])(X[n]-c[n]),m=1,2,∀,24,n=1,2,∀,24(12)(R=98.5%,)(R=97.3%),(R=88.6%)(R=64%),,(R=94.8%)(R=65.3%),(R=37.6%,33.3%),,2006,Qi[18][19](CSAC)OCTBarrett(BE),,82%,74%84%,OCT,,,[20],,,,(PCA),,OCT/,95%94%,OCT,100%,CSAC2007,LingleyPapadopoulos[21]KLaws#105,42K,,,59%,12%,17%,12%,,OCT[22][21],:,18,SW,,Si,Si=!j(xj-mi)(xj-mi)t,(13)mii,xjjSW=!iSi,(14)SB,SB=!ini(mi-m)(mi-m)t,(15)m,nji,∃leaveoneout%[23],,92%62%,OCT,,,97%,,,011701348,011701[18,20],(),3,,,,OCT,,(SVM)(ANN),,,2007,Zawadzki[24],,OCT,,(,),,,,,,,[25],40%,,,,,,,2008,Jorgensen[4]OCT,OCT14,∃leaveonesampleout%[26],,,,∃leaveonepatientout%OCT,,,73%81%2009,Baroni[27]OCTOCT5,73,0.1500020OCT10,10,,,,,2009,Sun[28],,,,,,OCT,,,[25],,,,,011701448,011701[29]OCTBarrett,OCT,,,,(LDA)[30]Barrett,(ANOVA)OCT,,LingleyPapadopoulos[31]OCT,OCT,,;,,BaylorOCT,,,,,,,22,,,,112,,,,,,,OCT,,,87%58%,,OCT,OCT,OCT11OCTTable1SamplesandresultsofrelatedarticlesinvolvedintextureanalysisforOCTimagesIssuestudiedLateralresolution/mClassificationmodelResultsRef./yearSkin/fat/normallung/abnormallung14minimumerrorrateBayesianclassificationmodelcorrectclassificationrates:skin(98.5%)versusfat(97.3%)normallung(88.6%)versusabnormallung(64%)fat(94.8%)versusnormallung(65.3%)versusskin(37.6%)[17]/2003Barrett#sEsophagus/classificationandregressiontreessensitivity95%,specificity94%for1image/biopsysitesensitivity100%,specificity100%for3images/biopsysite[20]/2006011701548,011701(thedifferencesweresmallandwouldnotaffectanalysisofthelayerasawhole),only12%segmentedinerror[21]/2007Bladdercancer&50Decisiontreesensitivity92%(increasedto97%ifpapillarygrowthswereexcludedfromcalculation)specificity62%(increasedto87%ifthecasesofinfiltrativeinflammationwereremovedfromthestudy)[22]/2008Retinallayers4SupportvectormachineTheretinalpigmentedepithelium(RPE),photoreceptorlayers&retinalnervefiberlayer(RNFL)aroundtheopticnervehead(ONH)canbesegmented.[24]/2007BasalCellCarcinomas&ActinicKeratosis20SupportvectormachineAccuracy73%forActinicKeratosis(AK)fromBasalCellCarcinomas(BCC)&81%forBCCfromAK[4]/2008Retinallayers/ArtificialneuralnetworksForInnerRetina(IR,includinginnerganglioncellsandinnerplexiformlayer):sensitivity87.0%,specificity71.4%,accuracy79.2%forOuterRetina(OR,includingouterplexiformandinnerphotoreceptorlayers):sensitivity70.5%,specificity74.9%,accuracy72.7%forBackground:sensitivity87.3%,specificity98.5%,accuracy79.2%[27]/2007Nevusflammeus8SupportvectormachineAccuracy99%[28]/2009Barrett#sEsophagus/LineardiscriminationanalysisMoredistinctdifferencebetweennormalandBarrett#sEsophaguscanbeshownwithultrahighresolutionOCT.[29]/2008Bladdercancer/WaveletanalysisSensitivity87%andspecificity58%[31]/20095,OCT,,,OCTOCT,,:1);2)()();3)OCT,,OCT,:1)OCT,,;2);3),OCT,,,OCT,,,,,,;,OCT,011701648,011701[J].Science,1991,254(5035):117811812A.F.Fercher,W.Drex