Integrating intratumoral and peritumoral features to predict tumor recurrence in intrahepatic cholangiocarcinoma

Previous studies have suggested that the intratumoral texture features may reflect the tumor recurrence risk in intrahepatic cholangiocarcinoma (ICC). The peritumoral features may be associated with the distribution of microsatellites. Therefore, integrating the imaging features based on intratumoral and peritumoral areas may provide more accurate predictions in tumor recurrence (both early and late recurrences) than the predictions conducted based on the intratumoral area only. This retrospective study included 209 ICC patients. We divided the patient population into two sub-groups according to the order of diagnosis time: a training cohort (159 patients) and an independent validation cohort (50 patients). The MR imaging features were quantified based on the intratumoral and peritumoral (3 and 5 mm) areas. The radiomics signatures, clinical factor-based models and combined radiomics-clinical models were developed to predict the tumor recurrence. The prediction performance was measured based on the validation cohort using the area under receiver operating characteristic curve (AUC) index. For the prediction of early recurrence, the combined radiomics-clinical model of intratumoral area with 5 mm peritumoral area showed the highest performance (0.852(95% confidence interval (CI), 0.724-0.937)). The AUC for the clinical factor-based model was 0.805(95%CI, 0.668-0.903). For the prediction of late recurrence, the radiomics signature of intratumoral area with 5 mm peritumoral area had the optimal performance with an AUC of 0.735(95%CI, 0.591-0.850). The clinical factor-based showed inferior performance (0.598(95%CI, 0.450-0.735)). For both early and late recurrences prediction, the optimal models were all constructed using imaging features extracted based on intratumoral and peritumoral areas together. These suggested the importance of involving the intratumoral and peritumoral areas in the radiomics studies.