CS-Unet (Optimizing Vision Transformers for Medical Image Segmentation)

The codes for the work "Optimizing Vision Transformers for Medical Image Segmentation"[https://ieeexplore.ieee.org/abstract/document/10096379], which is accepted by ICASSP 2023. This work continues to be improved through comparative learning, and you can refer to:

• MCSC which published by BMVC 2023

1. Prepare data

• The datasets we used are provided by TransUnet's authors. Please go to "./datasets/README.md" for details, or please send an Email to jienengchen01 AT gmail.com to request the preprocessed data. If you would like to use the preprocessed data, please use it for research purposes and do not redistribute it (following the TransUnet's License).

2. Environment

• Please prepare an environment with python=3.7, and then use the command "pip install -r requirements.txt" for the dependencies.

3. Train/Test

• Run the train script on synapse dataset. The batch size we used is 24.

• Train

sh my_train.sh 

• Test

sh my_test.sh 

References

• TransUnet
• SwinTransformer
• SwinUnet

Citation

@inproceedings{liu2023optimizing, title={Optimizing Vision Transformers for Medical Image Segmentation}, author={Liu, Qianying and Kaul, Chaitanya and Wang, Jun and Anagnostopoulos, Christos and Murray-Smith, Roderick and Deligianni, Fani}, booktitle={ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, pages={1--5}, year={2023}, organization={IEEE} }