MUnet-Lite: A Mamba-Based Lightweight Network for Efficient Abdominal Image Segmentation

Ziyang Hou, Zhe Zhao, Yusong Lin
Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, PMLR 278:119-124, 2025.

Abstract

The human abdomen houses multiple vital organs, and medical imaging technology precisely captures pathological features, providing a foundation for clinical diagnosis and treatment. High-precision abdominal image segmentation is crucial for lesion localization, organ measurement, and surgical planning. However, existing methods face challenges in local feature extraction and multi-scale information modeling. To overcome the limitations of Transformer-based approaches, such as insufficient local information perception, large model size, and high computational cost, we propose MUnet-Lite, a lightweight segmentation model. It combines the Mamba method with a U-Net architecture, incorporating a residual spatial modeling unit for enhanced feature extraction and an efficient decoding unit to reduce computation. Experiments on the Synapse dataset show that MUnet-Lite achieves a Dice score of 83.79% and a Hausdorff distance of 16.43mm, with only 26.71M parameters and 925.9 GFLOPs, significantly lowering computational cost while maintaining high segmentation accuracy. This provides a practical solution for real-world applications.

Cite this Paper

BibTeX
@InProceedings{pmlr-v278-hou25a, title = {MUnet-Lite: A Mamba-Based Lightweight Network for Efficient Abdominal Image Segmentation}, author = {Hou, Ziyang and Zhao, Zhe and Lin, Yusong}, booktitle = {Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing}, pages = {119--124}, year = {2025}, editor = {Zeng, Nianyin and Pachori, Ram Bilas and Wang, Dongshu}, volume = {278}, series = {Proceedings of Machine Learning Research}, month = {25--27 Apr}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v278/main/assets/hou25a/hou25a.pdf}, url = {https://proceedings.mlr.press/v278/hou25a.html}, abstract = {The human abdomen houses multiple vital organs, and medical imaging technology precisely captures pathological features, providing a foundation for clinical diagnosis and treatment. High-precision abdominal image segmentation is crucial for lesion localization, organ measurement, and surgical planning. However, existing methods face challenges in local feature extraction and multi-scale information modeling. To overcome the limitations of Transformer-based approaches, such as insufficient local information perception, large model size, and high computational cost, we propose MUnet-Lite, a lightweight segmentation model. It combines the Mamba method with a U-Net architecture, incorporating a residual spatial modeling unit for enhanced feature extraction and an efficient decoding unit to reduce computation. Experiments on the Synapse dataset show that MUnet-Lite achieves a Dice score of 83.79% and a Hausdorff distance of 16.43mm, with only 26.71M parameters and 925.9 GFLOPs, significantly lowering computational cost while maintaining high segmentation accuracy. This provides a practical solution for real-world applications.} }
Endnote
%0 Conference Paper %T MUnet-Lite: A Mamba-Based Lightweight Network for Efficient Abdominal Image Segmentation %A Ziyang Hou %A Zhe Zhao %A Yusong Lin %B Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing %C Proceedings of Machine Learning Research %D 2025 %E Nianyin Zeng %E Ram Bilas Pachori %E Dongshu Wang %F pmlr-v278-hou25a %I PMLR %P 119--124 %U https://proceedings.mlr.press/v278/hou25a.html %V 278 %X The human abdomen houses multiple vital organs, and medical imaging technology precisely captures pathological features, providing a foundation for clinical diagnosis and treatment. High-precision abdominal image segmentation is crucial for lesion localization, organ measurement, and surgical planning. However, existing methods face challenges in local feature extraction and multi-scale information modeling. To overcome the limitations of Transformer-based approaches, such as insufficient local information perception, large model size, and high computational cost, we propose MUnet-Lite, a lightweight segmentation model. It combines the Mamba method with a U-Net architecture, incorporating a residual spatial modeling unit for enhanced feature extraction and an efficient decoding unit to reduce computation. Experiments on the Synapse dataset show that MUnet-Lite achieves a Dice score of 83.79% and a Hausdorff distance of 16.43mm, with only 26.71M parameters and 925.9 GFLOPs, significantly lowering computational cost while maintaining high segmentation accuracy. This provides a practical solution for real-world applications.
APA
Hou, Z., Zhao, Z. & Lin, Y.. (2025). MUnet-Lite: A Mamba-Based Lightweight Network for Efficient Abdominal Image Segmentation. Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, in Proceedings of Machine Learning Research 278:119-124 Available from https://proceedings.mlr.press/v278/hou25a.html.

Related Material