Gl-MambaNet: A global-local hybrid Mamba network for medical image segmentation

Xiaoyan Kui, Shen Jiang, Qinsong Li, Yifei Peng, Zhipeng Hu, Beiji Zou

Abstract: In recent years, deep learning has made significant progress in medical image segmentation, but it still faces challenges in capturing global features and managing high computational complexity. The Mamba architecture has garnered considerable attention in medical image segmentation tasks due to its ability to capture global information with low computational overhead while maintaining high segmentation accuracy. Based on this, we propose a dual-path network design that combines CNN and Mamba architectures, utilizing a global-local feature fusion module (GLFFM), a multi-stage attention enhancement module (MSAEM), and a semantic feature alignment enhancement module (SFAEM) to significantly improve semantic consistency and multi-scale feature fusion. Specifically, GLFFM is employed to integrate the complementary features of CNN and Mamba, MSAEM enhances feature representation through channel and spatial attention mechanisms, while SFAEM precisely activates relevant features using group attention mechanisms and ensures consistency in dimensions and resolution across skip connection feature maps. Experimental validation on three publicly available medical image datasets, namely GlaS, PH2, and DSB2018, demonstrates that the proposed model, GL-MambaNet, outperforms existing state-of-the-art methods in terms of segmentation accuracy, memory usage, and computational efficiency.

External IDs:dblp:journals/ijon/KuiJLPHZ25