Abstract Precise delineation of hepatic and portal venous anatomy is crucial for the diagnosis of liver disease, surgical planning, and prognosis prediction. Current three-dimensional visualization of these complex vascular structures relies on manual or semi-automated CT segmentation, which is time-consuming and operator-dependent. Although artificial intelligence (AI) presents a promising alternative, existing methods remain constrained by the scarcity of publicly available datasets with fine-grained vascular annotations and inadequate validation in real-world diseased liver populations, which represent the majority of patients undergoing hepatic procedures. To address this gap, we present the Hepatic Vessel Map (HVM) Dataset, a dual-center resource comprising contrast-enhanced CT scans from 282 patients with over 4,1400 slices and 4,8300 annotations, each with meticulously annotated hepatic veins, portal veins (to third-order branches), and liver tumors. The dataset comprises a substantial proportion of cases with underlying hepatic pathology and has been validated for use in preoperative planning for major hepatectomy, ensuring both clinical relevance and model generalizability. This dataset supports: 1) development and benchmarking of robust hepatic and portal venous segmentation models; 2) vasoimcs research through quantitative analysis of vascular morphology, topology, and radiomic features; 3) generation of patient-specific 3D “digital vascular roadmaps” to enhance surgical precision and safety. As such, this dataset establishes a foundational resource for advancing AI-driven innovations in hepatobiliary surgery and intervention.
Xie et al. (Tue,) studied this question.