A Multi-parameters Radiomic Prediction Model For Preoperative Early Recurrence Risk Evaluation In Pancreatic Cancer

Background:Pancreatic cancer is one of the most progressive malignancies,and nearly half of the patients experienced recurrence after surgery.Therefore,preoperative evaluation of risk of early recurrence is significantly important.In this study,we aim to develop a novel prediction model based on radiomics features,and help surgeons with clinical strategies.Method:From April 2012 to July 2018,303 patients were retrospectively enrolled in two medical centers.We divided patients into early-recurrence(ER)group(disease free survival ? 12 months)and Non-ER group(disease free survival>12 months).Patients from first center were included as training cohort(n=123 patients)and internal validation cohort(n=53 patients).Patients from the second center were enrolled as validation cohort(n=126 patients).LASSO regression was applied for data dimension reduction,after which a radiomic signature was built for risk evaluation.Multivariate analysis was used for identification of independent risk factor.A nomogram was developed based on optimal factors.C-index and calibration curve were used to assess model performance.DCA curve was used for clinical value evaluation.Results:The C-index for radiomic signature in training cohort,internal validation cohort and external validation cohort was 0.81,0.80 and 0.78 respectively.After multivariate regression analysis,radiomic signature,CA19-9 level and clinical stage were identified as independent risk factors for early recurrence.Then a nomogram incorporating these factors was developed.The C-index for radiomic nomogram in training cohort,internal validation cohort and external validation cohort was 0.85,0.88 and 0.85 respectively.DCA curve showed the radiomic nomogram has the highest clinical value for application.Conclusion:Radiomics showed potential value for preoperative risk evaluation in pancreatic cancer.Our model may provide surgeons and oncologists with useful information and might serve as reliable reference for clinical treatment strategies.