An Accurate and Interpretable Hyper Graph Neural Network for GBM Survival Prediction

A sheaf hypergraph neural network integrates multi-modal MRI and clinical metadata to predict glioblastoma multiforme survival. The architecture uses directional, asymmetric message passing to capture higher-order tissue patch relationships. A concept bottleneck layer enables simultaneous predictive accuracy and model interpretability by mapping latent features to human-understandable concepts.