Clinically deployable AI to predict objective response to radiotherapy-intensified immunotherapy in advanced hepatocellular carcinoma

Background: This study aimed to evaluate whether radiotherapy enhances outcomes in advanced hepatocellular carcinoma (HCC) treated with immune-targeted therapy and to develop an interpretable artificial intelligence model for predicting response. Methods: In this multicenter retrospective study, 238 patients with HCC receiving immune-targeted therapy across three hospitals were included and categorized into an RT group or a no-radiotherapy (No-RT) group according to whether RT was delivered during treatment. Propensity score matching (PSM) was applied to mitigate baseline imbalance. For objective response rate (ORR) prediction, patients were randomly split (7:3) into training and validation cohorts, and eight AI models were developed and evaluated. Results: ORR was higher in the RT group than in the No-RT group (43.3% vs 28.8%, P = 0.02). RT was associated with longer overall survival (OS) and progression-free survival (PFS) both before and after PSM (all P < 0.05). Responders exhibited markedly improved OS and PFS compared with non-responders (both P < 0.001). Among eight models, the multilayer perceptron (MLP) achieved the best discrimination in the validation cohort (AUC-ROC = 0.71). SHapley Additive exPlanations (SHAP) highlighted age, tumor size, alpha-fetoprotein (AFP), and RT status as the dominant contributors. Conclusions: In advanced HCC, adding RT to immune-targeted therapy was associated with improved response and survival. An interpretable MLP model may offer a feasible, clinic-friendly approach to ORR prediction and support individualized immunoradiotherapy decisions.