MIST: Multi-instance selective transformer for histopathological subtype prediction

Rongchang Zhao, Zijun Xi, Huanchi Liu, Xiangkun Jian, Jian Zhang, Zijian Zhang, Shuo Li

Abstract: Highlights•A novel multi-instance selective transformer is proposed for the first time to formulate histopathological subtype prediction as fine-grained representation learning.•A selective instance transformer (SiT) is proposed to learn the instance- level fine-grained representation in histopathological subtype prediction by selecting the representative instances with a self-attention learning paradigm.•A multiple instance feature decoupling (MIFD) is proposed to leverage information bottleneck into the fine-grained representation learning and conduct accurate histopathological subtype prediction.