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

Yuyang Zheng; Mingda Zhang; Jianglong Qin; Qi Mo; Jingdan Pan; Haozhe Hu; Hongyi Huang

cs.CVMar 20, 2026

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

Authors: Yuyang Zheng, Mingda Zhang, Jianglong Qin, Qi Mo, Jingdan Pan, Haozhe Hu, Hongyi Huang

Organizations: School of Software Engineering, Yunnan University, KunMing, 650500, Yunnan, China. · The School of Artificial Intelligence, The Chinese University of Hong Kong, Shenzhen, 518172, Guangdong, China. · Yunnan Provincial Key Laboratory of Software Engineering, Yunnan University, Kunming, 650500, Yunnan, China.

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 & tables

Figure 1: 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 $F_{out}$ .

Dataset	Method	mDice	mIoU	Params (M)
UW-Madison GI Tract	U-Net Ronneberger et al. (2015)	0.8312	0.7562	32.66
	Att-UNet Oktay et al. (2018)	0.8314	0.7564	67.85
	TransUNet Chen et al. (2024)	0.8163	0.7394	117.19
	Swin-UNet Cao et al. (2023)	0.8292	0.7531	41.48
	UNet++ Zhou et al. (2018)	0.8311	0.7487	92.63
	MSDUNet (2025) Li et al. (2025)	0.8325	0.7575	116.82

Table 1: Comparison with state-of-the-art methods on the UW-Madison GI Tract and WORD datasets. The best result in each group is shown in bold .

Dataset	Organ	Dice	IoU	HD95 (mm)	ASD (mm)
UW-Madison GI Tract	Stomach	0.8862	0.8421	6.83	2.35
	Small Bowel	0.8252	0.7400	15.41	3.87
	Large Bowel	0.8735	0.7927	15.19	4.03
WORD	Liver	0.9660	0.9512	5.34	1.40
	Right Kidney	0.9751	0.9639	2.38	0.85
	Spleen	0.9807	0.9699	1.73	0.64

Table 2: Per-organ segmentation results of CS-MUNet on the UW-Madison GI Tract and WORD datasets. HD95 is reported in mm; ASD is reported in mm.

Dataset	Configuration	mDice	mIoU
UW-Madison GI Tract	Baseline	0.8312	0.7562
	Baseline + BASM	0.8432	0.7719
	Baseline + CMSA	0.8484	0.7755
	w/o $\Delta$ /B modulation	0.8537	0.7819
	w/o SE-style weight allocation	0.8484	0.7749
	CS-MUNet (Ours)	0.8616	0.7916

Table 3: Ablation study of CS-MUNet on the UW-Madison GI Tract and WORD datasets. The best result in each group is shown in bold .

Experiment	Value	mDice (%)	mIoU (%)	Optimal
Vary $\mu_{R}$ (fixed $\mu_{E}\!=\!1.2$ )	0.5	85.64	78.58	$\times$
	0.8	86.16	79.16	$\checkmark$
	1.2	85.52	78.39	$\times$
Vary $\mu_{E}$ (fixed $\mu_{R}\!=\!0.8$ )	0.6	85.55	78.47	$\times$
	1.2	86.16	79.16	$\checkmark$
	1.5	85.69	78.60	$\times$

Table 4: Hyperparameter sensitivity analysis on the UW-Madison GI Tract dataset. All metrics are reported as percentages (%). Bold indicates the optimal value in each group.

Figure 2: Comparison of parameter count (M) and computational cost (FLOPs, G) across all comparison models and ablation variants, with Ours highlighted in red.

Figure 3: Qualitative segmentation comparisons across five representative WORD CT samples (coronal, sagittal, and axial views).

Figure 4: Per-slice segmentation comparisons on five representative UW-Madison MRI cases across three organ categories: stomach (red), small bowel (green), and large bowel (blue).

Medical Image Segmentation

Medical image segmentation delineates organs, lesions, and other structures in scans to support measurement, diagnosis, treatment planning, and research.

See all 786 papers

MD-RWKV-UNet: Scale-Aware Anatomical Encoding with Cross-Stage Fusion for Multi-Organ Segmentation
Mar 28, 2026Zhuoyi FangMulti-Organ Segmentation Medical Image Segmentation
School of Mathematical Sciences Yangzhou University
CSG-Mamba: A Convolutional Scoring Gating Vision State Space Network for Endoscopic Polyp Segmentation
Aug 14, 2026Yuliang Wang, Jiaqi Wu, Jiaye Song +1Polyp Segmentation Medical Image Segmentation
School of Advanced Interdisciplinary Studies, Tiangong University, Tianjin, China
Multi-planar 2D-U-Net Segmentation of 3D-CT Abdominal Organs augmented by Spatial Occurrence Maps
Jun 5, 2026Daria Kern, Negar Chabi, Souraj Adhikary +1Multi-Organ Segmentation U-Net
Glasgow Caledonian University School of Science & Engineering Glasgow G4 0BA, United Kingdom · Jade University of Applied Sciences Department of Engineering & Medical Technology Germany, 26389, Wilhelmshaven, Germany

Mar 28, 2026J/K move · Enter open · S save

MD-RWKV-UNet: Scale-Aware Anatomical Encoding with Cross-Stage Fusion for Multi-Organ Segmentation

Zhuoyi Fang

School of Mathematical Sciences Yangzhou University

Multi-Organ Segmentation Medical Image Segmentation Multi-Scale Feature Fusion Multi-Scale Attention

Open paper AskPDF arXiv

Multi-organ segmentation in medical imaging remains challenging due to large anatomical variability, complex inter-organ dependencies, and diverse organ scales and shapes. Conventional encoder-decoder architectures often struggle to capture both fine-grained local details and long-range context, which are crucial for accurate delineation - especially for small or deformable organs. To address these limitations, we propose MD-RWKV-UNet, a dynamic encoder network that enables scale-aware representation and spatially adaptive context modeling. At its core is the MD-RWKV block, a dual-path module that integrates deformable spatial shifts with the Receptance Weighted Key Value mechanism, allowing the receptive field to adapt dynamically to local structural cues. We further incorporate Selective Kernel Attention to enable adaptive selection of convolutional kernels with varying receptive fields, enhancing multi-scale interaction and improving robustness to organ size and shape variation. In parallel, a cross-stage dual-attention fusion strategy aggregates multi-level features across the encoder, preserving low-level structure while enhancing semantic consistency. Unlike methods that stack static convolutions or rely heavily on global attention, our approach provides a lightweight yet expressive solution for dynamic organ modeling. Experiments on Synapse and ACDC demonstrate state-of-the-art performance, particularly in boundary precision and small-organ segmentation.

Abstract

Medical Image Segmentation

Explore similar work

MD-RWKV-UNet: Scale-Aware Anatomical Encoding with Cross-Stage Fusion for Multi-Organ Segmentation

CSG-Mamba: A Convolutional Scoring Gating Vision State Space Network for Endoscopic Polyp Segmentation

Multi-planar 2D-U-Net Segmentation of 3D-CT Abdominal Organs augmented by Spatial Occurrence Maps