This is the official PyTorch implementation of our paper, coined:

Learning Robust Shape Regularization for Generalizable Medical Image Segmentation.

If you have any problems, please email me (ck.ee@my.cityu.edu.hk).

please see the requirement file

1. download the Fundus dataset at https://drive.google.com/file/d/1zTeTiTA5CBKOCPq_xVRajWVKUtjjPSrF/view?usp=sharing and put it into the dir at "your_path/fundus/*"
2. create the environment as required by the requirement file
3. set key training parameters for train.py file:
   • "datasetTrain" - source domains, such as [1,2,4]
   • "datasetTest" - target domain, such as [3]
   • "data-dir" - where your dataset as step 1
   • "label" - determine the objective ( OC or OD ?) of validation for best model choice.
4. run train.py and you will get the saved model

1. set the you wanted model dir in test_visulization.py
2. run test_visualization.py

We offered a .ckpt file at https://drive.google.com/file/d/1-ntNwztBANmKnkf6VBZqEWGPqYL5sWg-/view?usp=sharing (OD - Target domain=4 - ASD=0.831 - Dice=0.936 ).

You can use this model to get a fast testing process on the OD #4 target domain using test_visualization.py

If you find our code and paper useful for you, please cite:

@article{ck2024tmi,
  title={Learning Robust Shape Regularization for Generalizable Medical Image Segmentation},
  author={Kecheng Chen, Tiexin Qin, Victor Ho-Fun Lee, Hong Yan, and Haoliang Li},
  journal={IEEE Transactions on Medical Imaging},
  year={2024},
  publisher={IEEE}
}

Many Thanks!