MambaMIM: Pre-training Mamba with state space token interpolation and its application to medical image segmentation

Fenghe Tang, Bingkun Nian, Yingtai Li, Zihang Jiang, Jie Yang, Wei Liu, S. Kevin Zhou

Abstract: Highlights•The first attempt to explore the potential of masked image modeling on Mamba.•A novel TOKI prioritizes consistency across the state space for causal learning.•A novel bottom-up strategy keeps mask consistency between different architectures.•Introducing hierarchical hybrid decoding to learn multi-scale representations.•Showing effectiveness on CT abdomen, lesion and MR tumor segmentation.