基于细胞图像肺癌诊断系统的设计与实现

66200712()JournalofJiangnanUniversity(NaturalScienceEdition)Vol.6No.6Dec.2007:1671-7147(2007)06-0739-05:2007-08-28;:2007-09-17.:(BE2006011).:(1986-),,,.3:(1972-),,,,,..Email:[e-mail],,,3(,210093):,.,.,,,,,,,deBoor2Cox,,,.:;;;;deBoor2Cox;:TP18:ADesignandImplementationofLungCancerDiagnosisSystemBasedonCellImagesGUOXiao2wen,ZHAOBo,ZHULiang,GAOYang3(NationalLaboratoryforNovelSoftwareTechnology,NanjingUniversity,Nanjing210093,China)Abstract:IncellimagebasedLungCancerDiagnosisSystem,DigitalImageProcessingandPatternRecognitionareusedtoprocessthelungcancerimages.Therefore,thusearlypathologicaldiagnosiscanbemadebasedonthefeaturesextractedfromtheselungcancerimages.Thepaperproposesanovelmethodtoextractcellareaandreconstructtheoverlappedones.Firstly,thecoloredimagesaretransformedtograyimagesandfiltered.Secondly,areinforcementlearningbasedmethodisusedtofindoptimumthresholdtosegmenttheimagesintoBi2levelimagesafterfilteringtheBi2levelimages.Besides,thepaperusesaBsplinebasedandimproveddeBoor2Coxmethodtodivideandreconstructtheoverlappedcells,thensomemorphologicalandcolorfeaturesareextractedfromthecellimagesandatlastthesefeaturesareusedtoclassifythelungcancercells.Keywords:reinforcementlearning;cellsegmentation;overlap;reconstruction;deBoor2Cox;featureextraction,.,,,,,,.[122],Agent,,,,,.,deBoor2Cox[3,4],,B,.,,[526].,weka.11.5,,.Agent,,deBoor2Cox.1Fig.1Systemarchitecture22.1,.RGB,,.,,,:H=0.71R+0.25G+0.05B(1),[6].2.2,[6].,,.,.22,Agent,,.Agent.,,Agent,.Agent.,:1)(S)EF:E=|ErEs||Es|(2)ErEsSobel[6].F=|FrFo||Fo|(3)FrFo(OSTU)[6]().EFS:S=EF(4)047()6EF1100,10000.2)(A)A=[-30-10-5-10151030](5)Tr,A[i]Tr:Tr=Tr+A[i](6)3)(Rw)Rw=|BrBn|+|FrFn||Bn+Fn|(7)BrBn,FrFn.Q2[7],Q2Q(s,a)SA.Q(s,a)0,Q(s,a).2greedy,,Tr,TrS,Q(s,a),1-Q(s,a)a,,r,(8)Q(s,a).Q(s,a)wQ(s,a)+[r+maxaQ(s,a)-Q(s,a)](8)ss,,.Q2Q(s,a)().Q(s,a),,Q(s,a),,.[6].2.3,BdeBoor2Cox.,deBoor2Cox.,().,,B,.,1/32/3,,3,,.deBoor2Cox.2,(ax,ay)(bx,by),(cx,cy),(ax,ay)(bx,by)(cx,cy)(dx,dy)(ex,ey),(dx,dy)(ex,ey)(fx,fy),(ax,ay)(dx,dy)(gx,gy),(bx,by)(ex,ey)(hx,hy),,,,.,B.2deBoor2CoxFig.2ImproveddeBoor2Coxalgorithm2.4,.1.[6].1476:..,.,,{R,G,B}{H,I,S}..,P,.1Tab.1FormulasforimportantfeaturesC=4SL2S,LM=SWHW,HE=min{W,H}max{W,H}f(k)=x(k)+jy(k)F()=1NN-1k=0f(k)exp(-j2k/N)x(k),y(k).u=mx=0ny=0(x-x)(y-y)f(x,y)f(x,y)(x,y)Gf=BR+G+Bf1=N1N1P(i,j)2f2=N1N1P(i,j)log(P(i,j))f3=maxi,j(P(i,j))f4=N1N1P(i,j)(i-j)233.13,,.(a);(b);(c)3Fig.3ThecompareofOSTUalgorithmandours3.24BdeBoor2Cox.,.,,.(a);(b);(c)(d)4BDeBoor2CoxFig.4Thecompareofwatershedalgorithmandours3.32.4,weka.[8],.2,.,,,.2Tab.2Precisionofclassification%91.272.48583.19488.977.951.247.378.288.273.171.871.471.180.494.579.4247()64,.,,,,.,,.,BdeBoor2Cox.,.,.,,.,,.:[1]KaelblingLP,LittmanML,MooreAW.Reinforcementlearning:Asurvey[J].JournalofArtificialIntelligenceResearch,1996,4(2):2372285.[2]LINing,GAOYang,LUXin,etal.Alearningagentbasedonreinforcementlearning[J].ComputerResearchandDevelopment,2001,38(9):105121056.[3]DonaldHearn,PaulineBakerM.ComputerGraphicswithOpenGL[M]//ThirdEdition.Beijing:PublishingHouseofElectronicsIndustry.2004.[4],.[M].:.1999.[5],,.[J].,2000,13(1):1162120.LUXing2quan,LINing,CHENShi2fu.Theresearchandimplementationofmorphologyandchromaindiagnosissystemforearlystagecomcer[J].RecognitionandArtficialIntelligence,2000,13(1):1162120.(inChinese)[6]CastlemanKR.[M]//.:,1998:4122453.[7]WatkinsJCH,DayanP.Q2learning[J].MachineLearning,1992(8):2792292.[8]HANJia2wei,MichelineKamber.[M]//,,.:,2001:2792335.(:)3476: