CS-MUNet: A Channel-Spatial Dual-Stream Mamba Network for Multi-Organ Segmentation

Abstract

Recently Mamba-based methods have shown promise in abdominal organ segmentation. However, existing approaches neglect cross-channel anatomical semantic collaboration and lack explicit boundary-aware feature fusion mechanisms. To address these limitations, we propose CS-MUNet with two purpose-built modules. The Boundary-Aware State Mamba module employs a Bayesian-attention framework to generate pixel-level boundary posterior maps, injected directly into Mamba's core scan parameters to embed boundary awareness into the SSM state transition mechanism, while dual-branch weight allocation enables complementary modulation between global and local structural representations. The Channel Mamba State Aggregation module redefines the channel dimension as the SSM sequence dimension to explicitly model cross-channel anatomical semantic collaboration in a data-driven manner. Experiments on two public benchmarks demonstrate that CS-MUNet consistently outperforms state-of-the-art methods across multiple metrics, establishing a new SSM modeling paradigm that jointly addresses channel semantic collaboration and boundary-aware feature fusion for abdominal multi-organ segmentation.

Figures (5)

• Figure 1: Comparative mDice performance of CS-MUNet against state-of-the-art methods on the UW-Madison GI Tract and WORD benchmarks. CS-MUNet outperforms all competing methods on both datasets, achieving gains of +2.91% and +1.03% in mDice over the strongest baseline on UW-Madison and WORD respectively, demonstrating consistent superiority across both MRI and CT modalities.
• Figure 2: Overall architecture of CS-MUNet and its two proposed modules.(a) U-shaped encoder-decoder with CMSA at the bottleneck and BASM at each skip connection fusing $F_e$ and decoder feature $F_d$.(b)CMSA processes grouped channel sequences via a shared SSM, where $\bar{A}_k$,$\gamma_k$ denote the state transition matrix and cumulative decay under Lipschitz constraints, and $\mu_k$,$h_{k+1}$, and $y_k$ the input, hidden state, and output of the k-th channel.(c) BASM injects boundary posterior $P_b$ derived from guidance map M and distance field $\tilde{d}$ into VMamba's $\Delta$ and B parameters to yield $F_{ssm}$, fused with $F_{safs}$ to produce $F_{out}$.
• Figure 3: Comparison of parameter count (M) and computational cost (FLOPs, G) across all comparison models and ablation variants, with Ours highlighted in red.
• Figure 4: Qualitative segmentation comparisons across five representative WORD CT samples (coronal, sagittal, and axial views).
• Figure 5: Per-slice segmentation comparisons on five representative UW-Madison MRI cases across three organ categories: stomach (red), small bowel (green), and large bowel (blue).