Spatially-Aware Mixture of Experts with Log-Logistic Survival Modeling for Whole-Slide Images

Accurate survival prediction from histopathology wholeslide images (WSIs) remains challenging due to gigapixel resolutions, spatial heterogeneity, and complex survival distributions. We introduce a comprehensive computational pathology framework that addresses these limitations through four synergistic innovations: (1) Quantile-Gated Patch Selection to dynamically identify prognostically relevant regions; (2) Graph-Guided Clustering that groups patches by spatial-morphological similarity to capture phenotypic diversity; (3) Hierarchical Context Attention to model both local tissue interactions and global slide-level context; and (4) an Expert-Driven Mixture of Log-Logistics module that flexibly models complex survival distributions. On large-scale TCGA cohorts, our method achieves state-of-the-art performance, with time-dependent concordance indices of 0.644 ± 0.059 on LUAD, 0.751 ± 0.037 on KIRC, and 0.752 ± 0.011 on BRCA—significantly outperforming both histology-only and multimodal benchmarks. The framework provides improved calibration and interpretability, advancing the potential of WSIs for personalized cancer prognosis.