Hyper-graph learning and its applications for medical image analysis

The hyper-graph, as a flexible structure, can naturally model the complex and high-order correlations among data due to its degree-free hyper-edges. Due to its superiority in modeling high-order correlations, hyper-graph learning has recently become more and more popular. Therefore much effort has been devoted to investigating the application of hyper-graph learning in medical image analysis. In this chapter, we begin with some fundamental concepts of hyper-graphs. Then we present the overall frameworks and implementations of hyper-graph neural networks. Finally, we discuss two typical applications of medical image analysis, i.e., identification of diseases with CT imaging and survival prediction on whole slide images.