ThegoalofthisprojectistoenrichSEERregistrydatawithhigh?qualitypopulation?based biospecimendataintheformofdigitalpathology,machinelearningbasedclassificationsand quantitativepathomicsfeaturesets.Wewillcreateawell?curatedrepositoryofhigh?quality digitizedpathologyimagesforsubjectswhosedataisbeingcollectedbytheregistries.These imageswillbeprocessedtoextractcomputationalfeaturesandestablishdeeplinkageswith registrydata,thusenablingthecreationofinformation?rich,populationcohortscontaining objectiveimagingandclinicalattributes.SpecificexamplesofdigitalPathologyderivedfeature setsincludequantificationoftumorinfiltratinglymphocytesandsegmentationand characterizationofcancerorstromalnuclei.Featureswillalsoincludespectralandspatial signaturesoftheunderlyingpathology.Thescientificpremiseforthisapproachstemsfrom increasingevidencethatinformationextractedfromdigitizedpathologyimages (pathomicfeatures)areaquantitativesurrogateofwhatisdescribedinapathologyreport.The importantdistinctionbeingthatthesefeaturesarequantitativeandreproducible,unlikehuman observationsthatarehighlyqualitativeandsubjecttoahighdegreeofinter?andintra?observer variability.Thisdatasetwillprovide,aunique,population?widetissuebasedviewofcancer, anddramaticallyaccelerateourunderstandingofthestagesofdiseaseprogression,cancer outcomes,andpredictandassesstherapeuticeffectiveness. ThisworkwillbecarriedoutincollaborationwiththreeSEERregistries.Wewillpartner withTheNewJerseyStateCancerRegistryduringthedevelopmentphaseoftheproject(UG3). Duringthevalidationphaseoftheproject(UH3),theGeorgiaandKentuckyStateCancer Registrieswilljointheproject.Theinfrastructurewillbedevelopedinclosecollaborationwith SEERregistriestoensureconsistencywithregistryprocesses,scalabilityandabilitysupport creationofpopulationcohortsthatspanmultipleregistries.Wewilldeployvisualanalytictools tofacilitatethecreationofpopulationcohortsforepidemiologicalstudies,toolstosupport visualizationoffeatureclustersandrelatedwhole?slideimageswhileprovidingadvanced algorithmsforconductingcontentbasedimageretrieval.Thescientificvalidationofthe proposedenvironmentwillbeundertakenthroughthreestudiesinProstateCancer,Lymphoma andNSCLC,ledbyinvestigatorsatthethreesites.
ThegoalofthisprojectistoenrichSEERcancerregistrydatawithhigh?qualitypopulation?based informationarisingfromdigitizedPathologyslides.Dataextracteddirectlyfromdigitized pathologyimages(Pathomicsdata)promisetoprovideinformationnotconsistentlyavailable fromPathologyreports.Thisdatasetwillprovide,aunique,population?widetissuebasedview ofcancer,anddramaticallyaccelerateourunderstandingofthestagesofdiseaseprogression, canceroutcomes,andpredictandassesstherapeuticeffectiveness.Thescientificvalidationof theproposedenvironmentwillbeundertakenthroughthreestudiesinProstateCancer, LymphomaandNSCLC,ledbyinvestigatorsatthethreesites.
Ren, Jian; Hacihaliloglu, Ilker; Singer, Eric A et al. (2018) Adversarial Domain Adaptation for Classification of Prostate Histopathology Whole-Slide Images. Med Image Comput Comput Assist Interv 11071:201-209 |