Multi-Scale Heterogeneity-Aware Hypergraph Representation for Histopathology Whole Slide Images

TL;DR

This work tackles survival prediction from histopathology whole slide images by modeling complex, multi-scale interactions among diverse tissue entities. It introduces a heterogeneity-aware hypergraph representation built atop a pairwise heterogeneous graph, coupled with a Heterogeneous HyperGraph Transformer (H2GT) that uses HMHA and type-specific projections to propagate information. Across BRCA, BLCA, and GBMLGG cohorts from TCGA, the method achieves state-of-the-art C-index scores and demonstrates robust patient stratification via Kaplan-Meier analysis. The approach offers a scalable framework for leveraging cross-scale biological interactions in prognostic tasks and provides publicly available code.

Abstract

Survival prediction is a complex ordinal regression task that aims to predict the survival coefficient ranking among a cohort of patients, typically achieved by analyzing patients' whole slide images. Existing deep learning approaches mainly adopt multiple instance learning or graph neural networks under weak supervision. Most of them are unable to uncover the diverse interactions between different types of biological entities(\textit{e.g.}, cell cluster and tissue block) across multiple scales, while such interactions are crucial for patient survival prediction. In light of this, we propose a novel multi-scale heterogeneity-aware hypergraph representation framework. Specifically, our framework first constructs a multi-scale heterogeneity-aware hypergraph and assigns each node with its biological entity type. It then mines diverse interactions between nodes on the graph structure to obtain a global representation. Experimental results demonstrate that our method outperforms state-of-the-art approaches on three benchmark datasets. Code is publicly available at \href{https://github.com/Hanminghao/H2GT}{https://github.com/Hanminghao/H2GT}.

Figures (6)

• Figure 1: Left: Previous Hypergraphs only contain one type of node and hyperedge. Right: The Heteorgeneity-Aware HyperGraphs contain types of nodes and hyperedges.
• Figure 2: The overflow of our proposed multi-scale heterogeneity-aware hypergraph representation framework, which includes heterogeneous graph construction, heteorgeneity-aware hypergraph construction, and Heterogeneous HyperGraph Transformer (H2GT).
• Figure 3: An example of a Heterogeneous Multi-Head Attention (HMHA) node embedding update module, where node $v$ is associated with $e^1(v)$ and $e^2(v)$, those hyperedges belong to two distinct categories.
• Figure 4: Kaplan-Meier Analysis on three datasets, where patient stratifications of low risk (blue) and high risk (orange) are presented. Shaded areas refer to the confidence intervals. P-value $<$ 0.05 means the significant statistical difference in two groups, and the lower P-value is better.
• Figure 5: We constructed three Boundary Hyperedges for each WSI according to the distance of each node from the central point $\mathbb{C}$, namely Hyperedge $\mathbb{B} ^1$, Hyperedge $\mathbb{B} ^{\frac{2}{3}}$ and Hyperedge $\mathbb{B} ^{\frac{1}{3}}$.