Abstract
BACKGROUND: Ki-67-driven tumor proliferation influences treatment and prognosis in intrahepatic mass-forming cholangiocarcinoma (IMCC), but preoperative imaging biomarkers for Ki-67 assessment are lacking. This study explores the relationship between Magnetic resonance imaging (MRI) tumor-halo radiomic features, imaging features, and Ki-67 expression levels. METHODS: The preoperative clinical and MRI data of 151 patients with IMCC across 4 centers were analyzed in this retrospective study. The data from 108 patients across 3 centers were combined for training and validation, whereas 43 patients from the fourth center constituted the test cohort. Tumor regions of interests (ROIs) were delineated on axial images and expanded outward by 3, 5, 10, 15, and 20 mm to obtain the peritumoral ROIs with varying ranges. Features were extracted using PyRadiomics and selected through SelectFromModel combined with random forest. Models were built and validated using a K-nearest neighbor classifier and fivefold cross-validation. RESULTS: Age, lesion location, vascular involvement, suspicious lymph nodes, and radiologically assessed signal characteristics on T2-weighted imaging (T2WI) and Diffusion-weighted imaging (DWI) were identified as the independent predictors of Ki-67 expression in IMCC. Ultimately, the 3 mm multimodal combined model exhibited the highest predictive performance, with an area under the curve (AUC) of 0.990, 0.980, and 0.869 in the training, validation, and test cohorts, respectively. CONCLUSIONS: A combined model integrating radiomics (intra/peri-tumoral) and imaging features from biparametric MRI effectively predicts preoperative Ki-67 expression in IMCC.