Artificial intelligence-based vascular pattern profiling predicts prognosis and therapeutic response in...

Hepatocellular carcinoma (HCC) is characterized by active angiogenesis and heterogeneous vascular patterns. However, vascular pattern profiling in tumors and its clinical significance remain unexplored.

This study aimed to develop an artificial intelligence-based system for quantitative vascular pattern profiling in HCC and to investigate the associations between vascular patterns, prognosis, and therapeutic response.

We collected 1,201 CD34-stained whole-slide images from 971 HCC patients across four medical centers, and developed a HCC Vascular Pattern Analysis System (HCC-VPAS), which comprised a tumor region detector, a vessel segmentation network and a vessel subtype classifier. Using HCC-VPAS, vessel subtypes were identified and quantified, followed by vascular pattern classification and integrative transcriptomic analyses.

HCC-VPAS first characterized four vessel subtypes, including linear vessels (LV), branched vessels (BV), dilated vessels (DV), and vessels encapsulating tumor clusters (VETC). Subsequent quantification of these vessel subtypes defined four reproducible vascular patterns (VPs): VP-Quiescent, VP-Capillary, VP-Dilated, and VP-Encapsulating. Furthermore, HCC-VPAS revealed significant differences in time to recurrence (TTR) among the four vascular patterns, specifically, patients in the VP-Capillary group showed the longest TTR, and those in the VP-Encapsulating group showed the shortest TTR. Based on these differences, patients were grouped into a binary VP group (VPG): L-VPG, composed of VP-Quiescent and VP-Capillary, showed longer TTR, while S-VPG consisted of VP-Dilated and VP-Encapsulating and had shorter TTR. By combining transcriptome with VPs, we identified VP-associated genes (VAG) that gradually increased from VP-Capillary, VP-Quiescent, VP-Dilated to VP-Encapsulating HCC, and computed a composite VAG score that could predict patients' TTR, overall survival (OS) and response to anti-angiogenesis and immunotherapy regimens.

HCC-VPAS provides a fully automatic and reproducible framework for quantitative vascular pattern profiling, aiming to offer a clinically actionable tool for identifying patients at high risk of recurrence and guiding personalized therapeutic strategies.