DIPathMamba: A domain-incremental weakly supervised state space model for pathology image segmentation

Jiansong Fan, Qi Sun, Yicheng Di, Jiayu Bao, Tianxu Lv, Yuan Liu, Xiaoyun Hu, Lihua Li, Xiaobin Cui, Xiang Pan

Abstract: Highlights•Supervised models need pixel labels, static models struggle with multi-domain data.•DIPathMamba for multi-domain pathology segmentation using only image-level labels.•Introducing Contrastive Mamba Block to explore dual-granularity features in pathology.•Designing a Domain Parameter Constraint Model to share info and constrain parameters.•Proposing CIDSL to utilize limited labels and guide parameter learning in domains.