Development and Validation of an Interpretable Machine Learning Model for Predicting Tumor Recurrence After Microwave Ablation in Small Hepatocellular Carcinoma

Yongfen Ma, Lei Zhao, Qing Zhao Department of Gastroenterology, Zibo Central Hospital, Zibo, 255000, People’s Republic of ChinaThese authors contributed equally to this workCorrespondence: Qing Zhao, Department of Gastroenterology, Zibo Central Hospital, No. 54, Youth League West Road, Zhangdian District, Zibo, Shandong, 255000, People’s Republic of China, Tel +86-18560292715, Email qingzh0801@163.comObjective: Tumor recurrence remains a major clinical challenge following microwave ablation (MWA) for small hepatocellular carcinoma (sHCC). Accurate and interpretable prediction of post-ablation recurrence risk is essential for individualized surveillance and management strategies. This study aimed to develop and validate an interpretable machine learning (IML) model for predicting tumor recurrence after MWA in patients with sHCC.Methods: This retrospective study included 536 consecutive patients with sHCC treated with ultrasound-guided MWA between July 2017 and July 2023. A two-stage feature selection strategy combining Boruta and LASSO regression was applied to identify the most robust predictors. Six ML algorithms were developed and compared, including support vector machine (SVM), random forest (RF), k-nearest neighbors (KNN), decision tree (DT), extreme gradient boosting (XGBoost), and logistic regression (LR). Model discrimination, calibration, and clinical utility were evaluated using receiver operating characteristic (ROC) analysis, calibration curves, decision curve analysis (DCA), and confusion matrices. Model interpretability was assessed using SHapley Additive exPlanations (SHAP).Results: Tumor recurrence occurred in 29.1% of patients within 1 year after MWA. Seven variables, including maximum tumor diameter, Child-Pugh grade, cirrhosis, portal hypertension, platelet count (PLT), alpha-fetoprotein (AFP), and C-reactive protein (CRP), were selected as predictors. Among all models, XGBoost demonstrated the best performance, achieving an AUC of 0.889 in the validation set, with good calibration and favorable net clinical benefit on DCA. SHAP analysis provided transparent global and individualized explanations, identifying AFP and CRP as the most influential predictors and revealing nonlinear risk patterns.Conclusion: An interpretable XGBoost-based model using routinely available clinical variables accurately predicts recurrence after MWA in patients with sHCC. This model may serve as a clinically accessible approach for individualized risk stratification and post-ablation management; however, further external validation in multicenter cohorts is required before broader clinical application.Keywords: interpretable machine learning, small hepatocellular carcinoma, microwave ablation, tumor recurrence, SHapley Additive exPlanations