Machine learning integrating MRI and clinical features predicts early recurrence of hepatocellular carcinoma after resection

This study aims to construct a robust artificial intelligence (AI) model to predict early recurrence of hepatocellular carcinoma (HCC) following surgical resection, leveraging clinical blood biomarkers, pathological parameters, and MRI-derived features. We included 240 hepatectomy patients from two medical centers, collecting clinical blood biomarkers, MRI features, and postoperative pathological data. Feature reduction was conducted using Spearman correlation and the least absolute shrinkage and selection operator (LASSO) regression. Predictive models were constructed using five machine learning algorithms and validated on an external dataset. The models were subsequently compared. The ExtraTrees, XGBoost, and LightGBM models exhibited high predictive performance in the training set, with AUCs of 0.816 (95% CI 0.748-0.884), 0.978 (95% CI 0.958-0.998), and 0.898 (95% CI 0.846-0.950), respectively. In the validation set, their AUC values were 0.759 (95% CI 0.641-0.876), 0.789 (95% CI 0.684-0.894), and 0.760 (95% CI 0.650-0.869). Decision curve analysis indicated favorable net benefits for predicting early recurrence across all three models. Tumor margin and age were identified as significant factors, showing strong associations with early recurrence. This study developed AI model utilizing clinical blood biomarkers, MRI features, and pathological information to predict early recurrence of HCC after surgery. The models demonstrated good predictive performance and showed clinical applicability in predicting early recurrence, potentially assisting clinicians in identifying high-risk patients, guiding individualized surveillance, and optimizing postoperative management. However, inherent biases in this retrospective study necessitate further research for validation and refinement.