MRI影像组学对胰腺导管腺癌病理分化程度的预测价值

打开文本图片集

[Abstract]Objective：ToexplorethepreoperativevalueofMRIradiomicsmodelforthepathologicaldiferentiationdegree of pancreaticductaladenocarcinoma（PDAC）.Methods：ThepancreaticMRIimagesandclinicaldataof96PDACpatients wereretrospectivelycollected.Thepatientsweredividedintoamedium-highdiferentiationgroup（52cases）andalow diferentitiongroup（44cases）basedonthepathologicaldiferentiationdegreeaftersurgeryROIsweredelineatedforthe tumorlayerbylayer，fromtheTSPIR，DWI，arterial phase，parenchymal phaseandvenousphaseiamges.Theradiomics featureswere extractedusingtheopen-sourcesoftware，andaradiomicsmodelwasestablishedusingsupport vector machine（SVM）asaclassifier.96patientswererandomlydividedintoatrainingcohort（67cases）andatesting cohort（29cases）ataratioof7：3.Thenthefive-foldcros-validationwasusedtointerallyvalidatetheradiomicsmodel andidentifythebesthyperparametersinthetrainingcohort.ROCcurveswereusedtoevaluatetheeficacyofthemodel. Results：TheclinicaldataorMRIfeatureshadnosignificantdiferences betweenthemedium-highdiferentiationandlow differentiation groups（all P >0.05）.A radiomicsmodel named Mean-RFE-18-SVM，capable of distinguishing the pathological diferentiationdegreeofPDAC，wasestablishedusing18radiomicsfeaturesselectedfromfivesequenceimages.Themodel performedwellininternalvalidation，withanaverageAUCofO.829，andtheAUCsforthetrainingandtestingcohorts were0.933（95%CI0.874-0.991）and0.846（95%CI0.697-0.995），respectively，withtheacuraciesof0.881and0.828，the sensivitiesof10oOand0.875，thespecificitiesof0.742and0.769，respectivelyConclusions：Themodelestablishedbased onMRIradiomicsfeaturesdemonstratesgooddiagnosticeficacyinnon-invasivepreoperativepredictionofthepathological diferentiationdegreeofPDAC.Itcanprovidemoredecisionsupportinformationforpersonalizeddiagnosisandtreatmentas well as prognosis assessment of PDAC patients.

[Keywords]Pancreaticductaladenocarcinoma;Pathologicaldiferentiationdegree；Magneticresonanceimaging;Radiomics

胰腺导管腺癌（pancreatic ductal adenocarcinoma，PDAC）恶性程度高，预后差，5年生存率仅 10%[1-2] PDAC的病理分化程度是评估患者预后的重要指标，可指导临床制订治疗方案[3-4]。（剩余7442字）