Development of a radiomics nomogram based on the CE-CT features to predict the survival of up front resectable patients with pancreatic head cancer and suspected venous invasion

Abstract

Introduction: Pancreatic head adenocarcinoma is commonly diagnosed at an advanced stage when adjacent vascular invasion is present. This study aimed to establish a radiomics nomogram by integrating radiomics signature and clinical predictors to estimate overall survival(OS) in patients who underwent attempted curative resectional surgery for pancreatic head cancer with suspected peri-pancreatic venous invasion. Method: Data on patients with pancreatic head adenocarcinoma and suspected peripancreatic invasion who underwent pancreatic resection with venous reconstruction were retrospectively collected from 2012 to 2016 at two academic institutions. A total of 396 radiomics features were extracted from pretreatment CE-CT images of each patient. Least absolute shrinkage and selection operator (LASSO) regression was applied to select optimal features and generate a radiomics signature. The radiomics nomogram was developed by integrating the radiomics signature and clinical predictors. The performance of radiomics nomogram validated in the cohort of patients of second institution. Result: Radiomics signatures were significantly associated with pancreatic head adenocarcinoma patients' survival time. The radiomics nomogram combined with clinical predictors(CA-199 value, Peripancreatic venous abnormalities and Lymph node staging) provided good predictive accuracy of survival on calibration curves. The C-index of the model in predicting overall survival (OS) was 0.836 for the validation cohort. Conclusion: The nomogram accurately predicted OS in patients with pancreatic head cancer with suspected peripancreatic venous invasion after attempted curative pancreatic resectional surgery. These findings might aid clinicians with treatment decision-making and improve precise medicine.