GMIM: Self-supervised pre-training for 3D medical image segmentation with adaptive and hierarchical masked image modeling

Liangce Qi, Zhengang Jiang, Weili Shi, Feng Qu, Guanyuan Feng

Abstract: Highlights•We proposed a novel self-supervised framework to enhance medical segmentation tasks.•The proposed dynamic masking guide networks learn the boundaries of organs and tissues.•The proposed hierarchical masking improves the quality of the representations learned by the lower layers.•The proposed multiple pre-text tasks learn the semantic information and reduce the representation redundancy.