Abstract
BACKGROUND: This study aims to develop and validate predictive models for postoperative complications and early recurrence in hepatocellular carcinoma (HCC) patients. METHODS: In a 7:3 ratio, 633 post-hepatectomy HCC patients were modeled and validated. Using 1561 radiomic features from two-dimensional shear wave elastography (2D-SWE), predictive models for the Comprehensive Complication Index (CCI) and recurrence were constructed, refined by LASSO regression, and integrated with clinical and pathologic data. Model performance was assessed using calibration plots and AUCs. RESULTS: The CCI model included Rad signature (P < 0.001), platelet count (P = 0.012), neutrophil levels (P = 0.032), Emax (P = 0.023), and differentiation status (P = 0.045), with AUCs of 0.896 for the modeling cohort and 0.832 for validation. The recurrence model included Rad signature (P < 0.001), arginase (P = 0.038), patient age (P = 0.001), adjacent tissue's Emean (P = 0.009), and blood flow (P = 0.03), with AUCs of 0.854 and 0.844 for modeling and validation cohorts, respectively. The Rad signature AUC of 0.884 surpassed the nomogram's AUC of 0.854 in the modeling set. CONCLUSION: Radiomic-based CCI and recurrence models effectively predict complications and short-term recurrence in post-hepatectomy HCC patients, aiding in identifying high-risk individuals for timely intervention.