IMG-11. GUARD-HGG: An AI-driven decision system for personalized radiochemotherapy recommendation in high-grade glioma

Abstract We developed and validated a machine learning-based system, GUARD-HGG, to non-invasively predict radiochemotherapy (RCT) sensitivity in patients with high-grade glioma (HGG), with the goal of supporting personalized RCT recommendations. A retrospective training set of 1,221 patients (14,652 multi-sequence MRIs) was selected from 15,785 consecutively reviewed HGG cases at Beijing Tiantan Hospital. For external testing, 193 patients (2,316 MRIs) were selected from a pool of 5,397 HGG cases across five independent institutions: Shandong Provincial Hospital, Harbin Medical University Cancer Hospital, Sixth Medical Centre of the General Hospital of the People’s Liberation Army, Affiliated Hospital of Hebei University, and China-Japan Union Hospital of Jilin University. Additionally, a prospective test cohort of 274 patients (3,288 MRIs) was enrolled through the registered clinical trial NCT06454097. The GUARD-HGG model was constructed by integrating radiomic features extracted from preoperative MRIs using manual tumor segmentations with key clinical variables. The model achieved AUCs of 0.874, 0.866, 0.853, and 0.882 in the training, internal validation, external, and prospective datasets, respectively. SHAP analysis revealed that residual tumor, patient age, and several radiomic features, such as Busyness (which reflects local intensity variation), Maximum voxel intensity, Tumor shape elongation, and clusterProminence (indicating gray-level skewness), were important predictors of RCT sensitivity. The model-derived Radscore effectively stratified patients into prognostically distinct groups in survival analysis. Patients with a higher Radscore, which indicates greater predicted sensitivity to RCT, were found to have significantly improved overall survival. These findings suggest that the GUARD-HGG model provides accurate and interpretable predictions of RCT sensitivity and may serve as a clinically valuable tool for guiding personalized decision-making in HGG patients.