MambaU-Lite: A Lightweight Model based on Mamba and Integrated Channel-Spatial Attention for Skin Lesion Segmentation

Introduction

This repository includes the implementation of skin lesion segmentation on the ISIC-2018 and PH2 datasets, as introduced in the paper: "MambaU-Lite: A Lightweight Model based on Mamba and Integrated Channel-Spatial Attention for Skin Lesion Segmentation" https://arxiv.org/pdf/2412.01405.

Contributions

• Introducing a lightweight hybrid segmentation model that combines the strengths of both Mamba and CNN architectures, effectively leveraging their advantages to enhance performance while keeping computational costs reasonable.
• Building a novel sub-structure called P-Mamba was established and implemented to efficiently learn features of different levels.

Citation

If you find this reference implementation useful in your research, please consider citing:

@misc{nguyen2024mambaulitelightweightmodelbased,
      title={MambaU-Lite: A Lightweight Model based on Mamba and Integrated Channel-Spatial Attention for Skin Lesion Segmentation}, 
      author={Thi-Nhu-Quynh Nguyen and Quang-Huy Ho and Duy-Thai Nguyen and Hoang-Minh-Quang Le and Van-Truong Pham and Thi-Thao Tran},
      year={2024},
      eprint={2412.01405},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2412.01405}, 
}