Based on CT radiomics and clinical features to distinguish pathological subtypes of intrahepatic cholangiocarcinoma: A multicenter study

The large duct-type (LD-type) intrahepatic cholangiocarcinoma (ICC) demonstrates more aggressive behavior than the small duct-type (SD-type), necessitating precise preoperative identification. This study aimed to develop a machine learning-based nomogram integrating CT radiomics and clinical features for distinguishing LD-type ICC. This multicenter retrospective study (2014-2022) included 220 ICC patients(73 LD-type, 147 SD-type). Patients from Institution 1 were divided into training (n = 115) and test (n = 50) sets, with those from institution 2 and 3 constituting the external validation set(n = 55). Radiomics features from arterial and venous phase CT images and clinical predictors from multivariate analysis were used to build models with logistic regression (LR), multilayer perceptron (MLP), and support vector machine (SVM) algorithms. The optimal model was presented as a nomogram. Performance was assessed via the area under the curve (AUC) of the receiver operating characteristics(ROC), decision curve analysis(DCA) and calibration curve. Prognostic differences were assessed via Kaplan-Meier analysid. Twelve radiomics features (4 from the arterial phase and 8 from the venous phase) and two clinical factors (tumor location, intrahepatic bile duct dilation) were selected. The MLP-based nomogram demonstrated superior performance with AUCs of 0.972 in the training cohort, 0.926 in the test cohort, and 0.840 in the validation cohort.DCA demonstrated strong clinical applicability. LD-type showed significantly shorter 1-year disease-free survival versus SD-type. The developed nomogram effectively distinguishes ICC subtypes, supporting personalized treatment and prognostic evaluation.