Prediction Of Peripancreatic Vascular Invasion In Pancreatic Ductal Adenocarcinoma Based On Enhanced CT Radiomics

Objective: 1.To investigate the value of preoperative enhanced CT signs in evaluating peripancreatic vascular invasion of pancreatic ductal adenocarcinoma(PDAC).2.To explore the application value of nomogram model constructed by CT radiomics features combined with clinical features in predicting peripancreatic vascular invasion of PDAC before operation.Materials and methods: 1.The preoperative clinical,pathological and CT imaging data of 128 patients with PDAC confirmed by pathological biopsy who underwent surgical treatment in our hospital from 2015 to 2022 were retrospectively analyzed.According to whether the blood vessels were involved,they were divided into vascular invasion group and non-vascular invasion group.The clinical data of age,gender,preoperative carbohydrate antigen CA199,CA125,AFP and CEA were analyzed.The preoperative enhanced CT tumor location,tumor transverse diameter,maximum angle between tumor and surrounding blood vessels(T/artery angle,T/portal vein angle,T/superior mesenteric vein angle,etc.),peritumoral fat space and lymph node metastasis were observed.Subsequently,univariate and multivariate logistic regression were used to analyze the independent predictors of vascular invasion around PDAC in the two groups.For parameters with statistically significant differences,the receiver operating characteristic(ROC)curve was drawn,and the area under the curve(AUC)value was calculated to evaluate the ability of different parameters to predict vascular invasion.2.The preoperative clinical,pathological and CT imaging data of 101 patients with PDAC confirmed by surgery and pathology(43cases in vascular invasion group and 58 cases in non-vascular invasion group)were retrospectively collected.The included cases were randomly assigned to a training group(70 cases)and a validation group(31 cases).5622 radiomics features were extracted from arterial,venous and delayed CT images,and then Man Whitney U test,Select KBst and Least absolute shrinkage and selection operator(LASSO)were used to screen the radiomics features that can predict vascular invasion.Five different classifiers were selected to train and verify the prediction model.The accuracy(ACC),sensitivity(SEN)and specificity(SPE)of the model were calculated according to the confusion matrix.The ROC curve was used to evaluate the diagnostic efficacy of vascular invasion.The optimal omics model and clinical factors were combined to construct a nomogram,and the decision curve analysis(DCA)was used to evaluate the clinical practicability of the tumor vascular invasion prediction model.Results: 1.Among 128 cases of PDAC,there were 35 cases in vascular invasion group and 93 cases in non-vascular invasion group.There were significant differences in T/artery angle,T/portal vein angle,T/superior mesenteric vein angle,tumor diameter and CA199 between the two groups(P<0.05).Multivariate logistic regression results showed that T/arterial angle and T/portal vein angle were independent predictors of vascular invasion in PDAC patients.ROC curve analysis showed that the AUC values of T/arterial angle and T/portal vein angle in predicting PDAC vascular invasion were 0.642 and 0.687,respectively.2.The radiomics model based on three-phase enhanced CT images consists of 23 radiomics features.Among the radiomics prediction models established by five different classifiers,XGBOOST had the best diagnostic efficiency.The AUC,ACC,SEN and SPE of the training set for predicting vascular invasion were 0.92,0.957,0.966 and 0.88,respectively.The AUC,ACC,SEN and SPE of the validation set were 0.88,0.847,0.798 and 0.861,respectively.In addition,the diagnostic efficiency of the clinical-radiomics fusion model constructed by XGBOOST combined with clinical factors was further improved compared with the diagnostic efficiency of the single factor model.The AUC,ACC,SEN and SPE of the training set were 0.97,0.971,0.976 and 0.975,respectively.The AUC,ACC,SEN and SPE of the validation set were 0.92,0.932,0.886 and 0.982,respectively.The calibration curve results showed that the actual results of distinguishing PDAC vascular invasion were comparable to the prediction probability of the clinical-imaging fusion model,and the DCA showed that the fusion model could obtain the best net benefit value.Conclusion: 1.CT imaging signs can predict the expression of vascular invasion in PDAC before operation;T/artery angle and T/portal vein angle ≥180 °can indirectly evaluate vascular invasion.2.The clinical-radiomics fusion model established by the omics features extracted from the three-phase enhanced CT images combined with clinical factors has high accuracy and sensitivity in predicting PDAC vascular invasion before surgery,and can be used as a means of clinical treatment plan selection and evaluation of patient survival prognosis.