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

TL;DR

This work tackles the challenge of predicting cancer survival from gigapixel histopathology WSIs by introducing a four-component framework that jointly leverages spatial organization and heterogeneous tissue phenotypes. The approach combines Quantile-Gated Patch Selection to focus on prognostically relevant regions, Graph-Guided Clustering to capture phenotype diversity, Hierarchical Context Attention for local and global context, and an Expert-Driven Mixture of Log-Logistics to flexibly model multi-modal survival distributions, achieving state-of-the-art time-dependent concordance indices on LUAD, KIRC, and BRCA ($0.644$, $0.751$, $0.752$ respectively). Ablation studies show each component contributes to performance, with multi-expert EDMLL providing the best calibration and interpretability. The method demonstrates strong potential for interpretable, WSI-based prognostics in personalized cancer care, even when using histology alone compared with multimodal baselines.

Abstract

Accurate survival prediction from histopathology whole-slide images (WSIs) remains challenging due to their gigapixel resolution, strong spatial heterogeneity, and complex survival distributions. We introduce a comprehensive computational pathology framework that addresses these limitations through four complementary innovations: (1) Quantile-Gated Patch Selection for dynamically identifying prognostically relevant regions, (2) Graph-Guided Clustering to group patches by spatial and morphological similarity, (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. Across large TCGA cohorts, our method achieves state-of-the-art performance, yielding time-dependent concordance indices of 0.644 on LUAD, 0.751 on KIRC, and 0.752 on BRCA, consistently outperforming both histology-only and multimodal baselines. The framework further provides improved calibration and interpretability, advancing the use of WSIs for personalized cancer prognosis.

Figures (2)

• Figure 1: (a) Proposed survival framework. (b) Expert-Driven Mixture of Log-logistics block.
• Figure 2: Kaplan–Meier survival curves showing significant high- vs. low-risk separation across TCGA cohorts.