Abstract
OBJECTIVE: To develop a non-invasive model integrating clinical features with intra- and peritumoral pre-fusion radiomic features to predict portal hypertension (PHT) in hepatocellular carcinoma (HCC) patients. MATERIALS AND METHODS: This retrospective study included 884 HCC patients who underwent partial hepatectomy with intraoperative portal venous pressure measurement (January 2013-January 2020). Patients were randomly assigned to training (n = 707, 89 with PHT) and validation (n = 177, 23 with PHT) cohorts. Clinical predictors were identified using logistic regression. Radiomic features were extracted from intratumoral and peritumoral regions on portal venous phase CT images. Key features were selected using t-tests, correlation analysis, and LASSO regression. Following comparison of multiple feature sets via K-Nearest Neighbors, intra- and peritumoral pre-fusion radiomic features were selected. A logistic regression-based nomogram combining clinical predictors with this radiomic set was developed and compared with traditional models. RESULTS: Portal vein diameter (PVD), Child-Pugh score, and FIB-4 score were identified as independent risk factors for PHT. The combined clinical-radiomic model achieved superior predictive performance in both the training (AUC: 0.938, 95% CI: 0.918-0.959) and validation (AUC: 0.847, 95% CI: 0.760-0.935) cohorts. CONCLUSION: The clinical-only model outperformed all radiomics-based models in this study, suggesting that routinely available clinical parameters may provide a robust foundation for portal hypertension screening. This finding may serve as a reference for clinical resource allocation and indicates the potential existence of a simplified, cost-effective screening pathway without reliance on complex radiomic analysis. Notably, the combined clinical-radiomic model demonstrated superior predictive performance, highlighting the complementary value of radiomic features to clinical indicators to a certain extent. However, the incremental benefit should be carefully weighed against the added complexity and challenges of standardization. This study further suggests that portal vein diameter, Child-Pugh score, and FIB-4 score may serve as independent predictors, offering a preliminary reference for future exploration of non-invasive tools for portal hypertension prediction. External validation is still needed to further confirm the model's generalizability and clinical utility.