ESP-MedSAM: Efficient Self-Prompting SAM for Universal Domain-Generalized Medical Image Segmentation

📌 This is an official PyTorch implementation of ESP-MedSAM: Efficient Self-Prompting SAM for Universal Domain-Generalized Medical Image Segmentation

[arXiv] [BibTeX]

📰News

[2024.08.08] The pre-print paper has been uploaded!

[2024.08.07] Paper will be updated soon!

[2024.08.07] Code and model checkpoints are released!

🛠Setup

git clone https://github.com/xq141839/ESP-MedSAM.git
cd ESP-MedSAM
conda create -n ESP python=3.10
conda activate ESP
conda install pytorch==1.13.0 torchvision==0.14.0 torchaudio==0.13.0 pytorch-cuda=11.6 -c pytorch -c nvidia
pip install albumentations==0.5.2
pip install pytorch_lightning==1.1.1
pip install monai

Note: Please refer to requirements.txt

📚Data Preparation

The structure is as follows.

ESP-MedSAM
├── datasets
│   ├── image_1024
│     ├── ISIC_0000000.png
|     ├── ...
|   ├── mask_1024
│     ├── ISIC_0000000.png
|     ├── ...

🎪Segmentation Model Zoo

We provide all pre-trained models here.

MA-Backbone	MC	Checkpoints
TinyViT	Dermoscopy	Link
TinyViT	X-ray	Link
TinyViT	Fundus	Link
TinyViT	Colonoscopy	Link
TinyViT	Ultrasound	Link
TinyViT	Microscopy	Link

🎈Acknowledgements

Greatly appreciate the tremendous effort for the following projects!

• SAM
• TinyViT

📜Citation

If you find this work helpful for your project, please consider citing the following paper:

@article{xu2024esp,
  title={ESP-MedSAM: Efficient Self-Prompting SAM for Universal Domain-Generalized Medical Image Segmentation},
  author={Xu, Qing and Li, Jiaxuan and He, Xiangjian and Liu, Ziyu and Chen, Zhen and Duan, Wenting and Li, Chenxin and He, Maggie M and Tesema, Fiseha B and Cheah, Wooi P and others},
  journal={arXiv preprint arXiv:2407.14153},
  year={2024}
}