HGMSurvNet: A two-stage hypergraph learning network for multimodal cancer survival prediction

Area of Science:

• Oncology
• Biomedical Informatics
• Machine Learning

Background:

• Multimodal data integration (pathology, clinical, genomic) is crucial for cancer survival prediction.
• Challenges include extracting effective representations and handling missing data in clinical settings.

Purpose of the Study:

• To propose HGMSurvNet, a novel two-stage hypergraph learning network for multimodal cancer survival prediction.
• To address data noise and missing modalities in clinical data.

Main Methods:

• Developed a two-stage hypergraph learning network (HGMSurvNet) for gradual, higher-order representation learning.
• Implemented a hypergraph convolution network with a hyperedge dropout mechanism to manage noisy and missing data.

Main Results:

• HGMSurvNet consistently outperformed state-of-the-art methods across six TCGA cancer cohorts.
• Demonstrated interpretable analysis of HGMSurvNet's performance in pathological images and patient modeling.

Conclusions:

• HGMSurvNet offers a robust solution for multimodal cancer survival prediction, effectively handling missing data.
• The method shows significant potential for clinical application in survival prognosis and patient modeling.