This repository provides the code for our MICCAI 2025 paper "FDAS: Foundation Model Distillation and Anatomic Structure-aware Multi-task Learning for Self-Supervised Medical Image Segmentation"

1. Download the M&MS Dataset, and organize the dataset directory structure as follows.

your/data_root/ train/ img/ full/ # 100% of the training images A0S9V9_0.nii.gz ... subset20/ # 20% of the training images subset10/ # 10% of the training images ... lab/ full/ A0S9V9_0_gt.nii.gz ... subset20/ subset10/ ... valid/ img/ lab/ test/ img/ lab/ 

2. Download the SAM model and move the model to the "your_root/pretrained_model" directory in your project.

3. Generate masks using the Segment Anything Model (SAM)

generate_masks.py 

4. Pre-process the masks generated from SAM

pre_process.py 

5. Pre-train the network using pre_train.py

pre_train.py 

6. Fine-tune the network using fine_tune.py

fine_tune.py 

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

@inproceedings{qi2025fdas, title={FDAS: Foundation Model Distillation and Anatomic Structure-Aware Multi-task Learning for Self-Supervised Medical Image Segmentation}, author={Qi, Xiaoran and Zhang, Guoning and Wu, Jianghao and Zhang, Shaoting and Hou, Xiaorong and Wang, Guotai}, booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention}, pages={192--202}, year={2025}, organization={Springer} }

• Parts of our codebase are adapted from PyMIC. We also thank the authors of the Segment Anything project for open-sourcing their code.