Impact of vascular microenvironment on prognosis in hepatocellular carcinoma: A quantitative feature analysis.

e16293 Background: Hepatocellular carcinoma (HCC) is a leading cause of cancer-related mortality globally. The vascular component of the tumor microenvironment plays an important role in HCC progression, yet its prognostic implications remain to be fully elucidated. This study aims to investigate the prognostic value of tumor vascular related microenvironment in HCC by leveraging quantitative analysis to identify critical features that influence survival and serve as potential biomarkers. Methods: The TCGA-LIHC dataset was utilized, comprising 365 HCC patients with H&E stained whole slide images (WSI) and corresponding clinical data. After excluding samples with overall survival (OS) < 1 month or WSI artifacts, 336 cases were analyzed. The Pasegnet model classified tissue patches in WSI into tumor and non-tumor categories. Within tumor regions, the Omni-seg model was applied to perform blood vessel segmentation. Meanwhile, the CellVit model was used to segment cell nuclei, including tumor cells, inflammatory cells, and endothelial cells. A total of 225 features associated with the tumor vascular related microenvironment were extracted based on segmentations results, refined through statistical tests and random forests, and subsequently used to construct an XGBoost survival model for prognostic prediction. Results: The Pasegnet model applied to the above dataset achieved an accuracy (ACC) of 95.6% and an area under the curve (AUC) of 0.9984 for tissue classification. In vessel segmentation, the Omni-seg model obtained a DICE coefficient of 83.87% on validation set. For cell nuclear segmentation, the CellVit model demonstrated binary Panoptic Quality (bPQ) and mean Panoptic Quality (mPQ) values of 0.6345 and 0.5677 on validation set, respectively. Following feature selection, an XGBoost survival model achieved a C-index of 0.67 on the TCGA-LIHC validation set. Further analysis revealed 34 features significantly associated with prognosis. These features can be grouped into three categories: 11 morphological features (Vessel Radius, Vessel Area, etc.) , 6 structural features (Vessel Voronoi Polygon Area , Vessel Tortuosity, etc.), and 17 microenvironment distribution features (Count of Tumor Cells In Vessel, Vessel Density, etc.), collectively indicating tumor invasiveness, vascular abnormalities, and reflecting vascular structure and spatial organization. Conclusions: These results indicate that morphological, structural, and microenvironment distribution characteristics of the tumor vasculature are predictive of survival in HCC. Their identification underscores the tumor vasculature’s prognostic importance and may guide more personalized risk stratification and treatment planning.