A survival prediction network based on multi-scale hypergraph enhancement and cross-modal refinement
Abstract Cancer survival prediction, as a critical task in clinical prognosis analysis, holds significant importance for guiding treatment decisions and patient management. This paper proposes a survival prediction network based on multi-scale hypergraph enhancement and cross-modal refinement (MSHG-...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-08-01
|
| Series: | Journal of King Saud University: Computer and Information Sciences |
| Subjects: | |
| Online Access: | https://doi.org/10.1007/s44443-025-00202-3 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Cancer survival prediction, as a critical task in clinical prognosis analysis, holds significant importance for guiding treatment decisions and patient management. This paper proposes a survival prediction network based on multi-scale hypergraph enhancement and cross-modal refinement (MSHG-CMR), which achieves deep joint modeling and efficient information fusion of pathological images and multi-omics data by integrating multi-scale hypergraph enhancement and cross-modal refinement Transformer technology. In the method design, the hypergraph construction module captures high-order feature associations of pathological images at different resolutions and effectively aligns feature distributions across scales through homomorphic nonlinear mapping; subsequently, the cross-modal refinement Transformer module based on rotary encoding achieves dynamic information transfer and complementary refinement between visual features and multi-omics data through attention mechanisms and position decay designs; finally, a feature fusion attention mechanism is introduced to fuse multi-modal features, which both suppresses inter-modal redundancy and effectively mitigates information conflicts, thereby enhancing the precision of individualized prognosis analysis. Experimental results show that MSHG-CMR improves the C-index performance by 1.7% overall compared to existing SOTA methods on the TCGA-BLCA dataset and by 1.1% on the TCGA-LUAD dataset, demonstrating its generalizability across different cancer types. Code is available at: https://github.com/MUYI-XIAN/MSHG-CMR . |
|---|---|
| ISSN: | 1319-1578 2213-1248 |