Rethinking mitosis detection: Towards diverse data and feature representation for better domain generalization

Jiatai Lin, Hao Wang, Danyi Li, Jing Wang, Bingchao Zhao, Zhenwei Shi, Changhong Liang, Guoqiang Han, Li Liang, Zaiyi Liu, Chu Han

Abstract: Highlights•We propose a novel and generalizable mitosis detection framework (MitDet) that achieves SOTA performance on five popular datasets internally and externally comparing with existing models.•DGSB is proposed to balance the data quantity and diversity. InCDP is proposed to obtain a diverse feature representation. SE module is introduced to enhance the diversity of both data and features simultaneously.•Compared with existing approaches, including pixel-level and box-level ones, MitDet achieves state-of-the-art domain generalization performance using only point-level annotations.