Preoperative Prediction Of Pathological Grade Of Pancreatic Neuroendocrine Tumors Based On Contrast-enhanced CT Radiomics Model

Purpose: To explore the feasibility and value of a CT radiomics model in preoperative,noninvasive prediction of pathological grade of pancreatic neuroendocrine tumors(PNETs).Methods: One hundred and forty-five patients derived from two institutions with postoperative pathologically confirmed PNETs were included in this retrospective study,including 91 patients in the training cohort(Zhongshan Hospital affiliated to Fudan University,G1=34,G2/3=57)and 54 patients in the validation cohort(Guangdong Provincial people’s Hospital,G1=28,G2/3=26).The clinical and imaging data of the patients were collected retrospectively,and the radiomic features were extracted based on arterial phase(Arterial Phase,AP)and portal vein phase(Portal venous Phase,PP)CT images in the training cohort.Pearson correlation analysis and Relief F(Relevance In Estimating Features F)algorithm were adopted for the significant radiomic feature selection.Logistic regression was used to construct radiomics models to predict the pathological grade of PNETs based on the training cohort data.The diagnostic performance was analyzed with the receiver operating characteristic(ROC)curve and calculating the area under the curve,accuracy,sensitivity and specificity.Results: Combined radiomic model based on the arterial and portal phase CTimages in the training cohort showed good predictive performance.In the training cohort,the area under the ROC curve(AUC),accuracy,sensitivity and specificity were 0.86(95%CI:0.78-0.94),0.81,0.79 and 0.85,respectively.In the validation cohort,the AUC,accuracy,sensitivity and specificity were 0.85(95%CI:0.75-0.95),0.80,0.85 and 0.75,respectively.Conclusion: The radiomic model based on contrast-enhanced CT can be used as a non-invasive tool for preoperative prediction of grade 1 and grade 2/3 PNETs to guide the clinical decisionmaking.