Foundation Model-Enabled Pathology-Informed MRI Model for Preoperative Prediction of Vessels Encapsulating Tumor Clusters in Resectable Hepatocellular Carcinoma: A Multicenter Retrospective Study

Vessels encapsulating tumor clusters (VETC) are an aggressive vascular phenotype of hepatocellular carcinoma that is usually determined after surgery. A foundation model-enabled framework for pathology-informed VETC prediction on pretreatment magnetic resonance imaging (MRI), termed PIVOT, uses paired preoperative multiparametric MRI, postoperative hematoxylin and eosin whole-slide images, and CD34 immunohistochemistry during model development, while inference requires only preoperative MRI. PIVOT-MRI uses morphologic and vascular embedding alignment, and PIVOT-Clinical combines the PIVOT-MRI score with tumor size and alpha-fetoprotein. The multicenter retrospective study includes 2052 patients in a training cohort, an internal test cohort, and four external validation cohorts. PIVOT-MRI achieves an area under the receiver operating characteristic curve of 0.887 in the internal test cohort and 0.839-0.861 across external cohorts. PIVOT-Clinical increases the internal-test area under the curve to 0.904 and PIVOT-Clinical-defined groups are associated with different two-year recurrence risks. Removing both histopathologic alignment branches results in the largest ablation-related performance decrease. Exploratory radiotranscriptomic analyses showed that higher PIVOT-MRI scores were associated with angiogenesis-related, endothelial, and proliferative transcriptional programs. PIVOT enables noninvasive preoperative assessment of pathology-defined vascular aggressiveness in resectable hepatocellular carcinoma while preserving an MRI-only clinical inference pathway.