Inter-Intra Hypergraph Computation for Survival Prediction on Whole Slide Images

Area of Science:

• Computational pathology
• Medical image analysis
• Bioinformatics

Background:

• Histopathology whole slide images (WSIs) analysis for survival prediction requires understanding complex inter-patient and intra-image correlations.
• Current graph-based methods often overlook high-order correlations, limiting their representational capacity.
• Existing hypergraph methods struggle to unify multi-level high-order correlations.

Purpose of the Study:

• To propose a unified framework for integrating multi-level high-order correlations in WSIs for improved survival prediction.
• To address the limitations of existing graph and hypergraph methods in capturing complex correlations within gigapixel histopathology images.

Main Methods:

• Developed an inter-intra hypergraph computation (I$^{2}$2HGC) framework for multi-level hypergraph computation.
• Implemented intra-hypergraph computation to model high-order correlations among patches within individual WSIs.
• Employed inter-hypergraph computation using patient embeddings to model population-level high-order correlations, fusing intra- and inter-risks for final prediction.

Main Results:

• The hypergraph structure captures richer correlations than graph structures, including pairwise and higher-order interactions.
• The I$^{2}$2HGC framework effectively models topological and semantic information in WSIs.
• Experimental results on TCGA carcinoma datasets demonstrate superior performance compared to existing methods.

Conclusions:

• Hypergraph-based methods offer significant advantages for capturing complex correlations in large-scale medical image analysis, particularly for WSIs.
• The proposed I$^{2}$2HGC framework provides a powerful tool for survival prediction by integrating multi-level high-order correlations.
• This approach enhances the representation capability of WSIs for clinical outcome prediction.