Mamba 3D Medical Image Segmentation

This repository is a collection of mamba-based and U-shaped models tailored for the segmentation of medical images. It provides a collections of state-of-the-art models we used to compare with for different related publications.

Installation

Requirements: Ubuntu 20.04, CUDA 11.8

1. Create a virtual environment: conda create -n tamingmambas python=3.10 -y and conda activate tamingmambas
2. Install Pytorch 2.0.1: pip install torch==2.0.1 torchvision==0.15.2 --index-url https://download.pytorch.org/whl/cu118
3. Install Mamba: pip install causal-conv1d>=1.2.0 and pip install mamba-ssm --no-cache-dir
4. Download code: git clone https://github.com/LucaLumetti/TamingMambas.git
5. cd TamingMambas/tamingmambas and run pip install -e .

Sanity check: Enter the Python command-line interface and run:

import torch
import mamba_ssm

If you face problems with Mamba or causal-conv1d, try to install them manually.

Cite Us

If you find this project useful for your research or development, please consider citing it:

Accurate 3D Medical Image Segmentation with Mambas.

@inproceedings{lumetti2025accurate,
  title={Accurate 3D Medical Image Segmentation with Mambas},
  author={Lumetti, Luca and Pipoli, Vittorio and Marchesini, Kevin and Ficarra, Elisa and Grana, Costantino and Bolelli, Federico and others},
  booktitle={Proceedings of 2025 IEEE International Symposium on Biomedical Imaging (ISBI)},
  year={2025}
}

Taming Mambas for Voxel Level 3D Medical Image Segmentation.

@article{lumetti2024taming,
  title={Taming Mambas for Voxel Level 3D Medical Image Segmentation},
  author={Lumetti, Luca and Pipoli, Vittorio and Marchesini, Kevin and Ficarra, Elisa and Grana, Costantino and Bolelli, Federico},
  journal={arXiv preprint arXiv:2410.15496},
  year={2024}
}

Acknowledgements

We acknowledge all the authors of the employed public datasets, allowing the community to use these valuable resources for research purposes. We also thank the authors of nnU-Net, Mamba and U-Mamba for making their valuable code publicly available.