Predicting early recurrence after microwave ablation in hepatocellular carcinoma: a clinicopathological-radiomics model based on ultrasound and identification of minimum ablation margin for high-risk tumors.

AIM: To develop a comprehensive clinicopathologic-radiomic model for predicting early recurrence (ER) after thermal ablation for hepatocellular carcinoma (HCC), and to explore optimal ablation strategies for high-risk tumors. MATERIAL AND METHODS: This multicenter retrospective study of 325 HCC patients undergoing microwave ablation divided them into training (n=182), internal (n=78), and external test sets (n=65). We extracted 3,499 radiomic features from pre-procedural ultrasound images and combined them with clinicopathological variables to train seven machine learning classifiers using 10-fold cross-validation. Model performance was assessed primarily by the area under the ROC curve, with decision curve and calibration analyses for clinical utility. The optimal minimal ablative margin (MAM) threshold for high-risk tumors was determined in the internal set using ROC analysis and validated externally. RESULTS: The integrated clinicopathological-radiomics model demonstrated superior predictive performance, with an AUC of 0.870 (95% CI: 0.762-0.978) and showed good calibration and positive net benefit on decision curve analysis. SHapley Additive exPlanations analysis identified key predictive features, predominantly from arterial-phase CEUS images. In this cohort, a larger MAM (threshold 7.7 mm derived internally) was associated with lower ER in predicted high-risk tumors. CONCLUSION: An integrated clinicopathological-radiomics model effectively predicts ER of HCC following microwave ablation and provides an effective strategy for high-risk tumors to reduce ER occurrence.