Masked hypergraph learning for weakly supervised histopathology whole slide image classification

Jun Shi, Tong Shu, Kun Wu, Zhiguo Jiang, Liping Zheng, Wei Wang, Haibo Wu, Yushan Zheng

Abstract: Highlights•A novel weakly supervised framework named MaskHGL is proposed for histopathology WSI analysis.•Multiple Hypergraphs are built to exploit the intrinsic non-pairwise relationships in WSI.•MaskHGL enhances the representational ability of instances and alleviates the noise impact brought by instance-level pseudo-label.•MaskHGL achieves remarkable performance on cancer subtyping, gene mutation prediction, and fine-grained gene mutation site prediction.