MMHG: Multi-modal Hypergraph Learning for Overall Survival After D2 Gastrectomy for Gastric Cancer

Overall survival (OS) prediction has been a central topic of oncology. Most existing OS prediction models are based on traditional statistical methods which becomes a limitation when confronted with high dimensional dataset as well complicated internal relation among features in practice. In this paper, we weaken this limitation by exploring the application of multi-modal hypergraph (MMHG) learning framework to improve the accuracy of prediction. More specifically, the proposed hypergraph model unites the demographics, pathologic characteristics and physiological indicators simultaneously to predict the overall survival after D2 gastrectomy for gastric cancer. Experiments are carried out on a real data set of West China Hospital of Sichuan University with 939 patients to evaluate the proposed approach by comparison with random forest (RF) and support vector machine (SVM). Results demonstrate that our scheme outperforms the baseline methods in overall survival classification performance.