WJG:AnewendoscopicultrasonographyimageprocessingmethodtoevaluatetheprognosisforpancreaticcancertreatedwithinterstitialbrachytherapyAIM:Todevelopafuzzyclassificationmethodtoscorethetexturefeaturesofpancreaticcancerinendoscopicultrasonography(EUS)imagesandevaluateitsutilityinmakingprognosisjudgmentsforpatientswithunresectablepancreaticcancertreatedbyEUS-guidedinterstitialbrachytherapy.
METHODS:EUSimagesfromourretrospectivedatabasewereanalyzed.Theregionsofinterestweredrawn,andtexturefeatureswereextracted,selected,andscoredwithafuzzyclassificationmethodusingaC++program.Then,patientswithunresectablepancreaticcancerwereenrolledtoreceiveEUS-guidediodine125r
ADIoactiveseedimplantation.Theirfuzzyclassificationscores,tumorvolumes,andcarbohydrateantigen199(CA199)levelsbeforeandafterthebrachytherapywererecorded.Theassociationbetweenthechangesintheseparametersandoverallsurvivalwasanalyzedstatistically.
RESULTS:EUSimagesof153patientswithpancreaticcancerand63non-cancerpatientswereanalyzed.Atotalof25consecutivepatientswereenrolled,andtheytoleratedthebrachytherapywellwithoutanycomplications.Therewasacorrelationbetweenthechangeinthefuzzyclassificationscoreandoverallsurvival(Spearmantest,r=0.616,P=0.001),whereasnocorrelationwasfoundtobesignificantbetweenthechangeintumorvolume(P=0.663),CA199level(P=0.659),andoverallsurvival.Therewere15patientswithadecreaseintheirfuzzyclassificationscoreafterbrachytherapy,whereasthefuzzyclassificationscoreincreasedinanother10patients.Therewasasignificantdifferenceinoverallsurvivalbetweenthetwogroups(67dvs151d,P=0.001),butnotinthechangeoftumorvolumeandCA199level.
CONCLUSION:UsingthefuzzyclassificationmethodtoanalyzeEUSimagesofpancreaticcancerisfeas
IBLe,andthemethodcanbeusedtomakeprognosisjudgmentsforpatientswithunresectablepancreaticcancertreatedbyinterstitialbrachytherapy.
INTRODUCTION:Theapplicationofdigitalimageprocessing(DIP)inendoscopicultrasonography(EUS)imagesandotherimagingscenarioshasbeenproventobeausefuladjuncttoendoscopicdiagnosesandoftencomparablewithspecialists’interpretationindifferentpathologicsettings.ThetextureparametersofEUSimagesareextractedandclassifiedfromthereturnedechoestoidentifythetissuetypepresentintheimages.OneeffectiveapproachistouseDIPbasedonasupportvectormachine(SVM),whichisacomputeralgorithmthatlearnsbyexampletoassignlabelstoobjects.TheSVMtechnique,asasubfieldofdigitalsignalprocessing,hasbeenappliedtoaseriesofpathologicallyprovendiseases.
ThetypicalmethodofSVM,whichisonlyabletoprovideadifferentialdiagnosisforsolidtumors(“yes”or“no”),cannotprovidenumericaldatadescribingthetextureparametersintheEUSimage.Inthisstudy,anewDIPmethodbasedonfuzzyclassificationisappliedtoobtainthefeaturevalueoftextureparametersinEUSimagesofpancreaticcancerandobservethechangeoftextureparameterstoevaluateitsutilityinmakingprognosisjudgmentsforpatientswithunresectablepancreaticcancerafterEUS-guidedinterstitialbrachytherapy.