Simple Consistency Regularization for SSL-based Medical Image Segmentation

This is the official implementation of the paper titled "An Embarassingly Simple Consistency Regularization Method for Semi-Supervised Medical Image Segmentation" accepted in IEEE ISBI 2022.

Running the code

git clone https://github.com/hritam-98/ICT-MedSeg

Requirements

Some important required packages include:

• Pytorch version >=0.4.1.
• TensorBoardX
• Python == 3.6
• Efficientnet-Pytorch pip install efficientnet_pytorch
• Some basic python packages such as Numpy, Scikit-image, SimpleITK, Scipy ......

Dataset

Download the processed data and put the data in ../data/MMWHS or ../data/ACDC, please read and follow the ACDC.md.

Training

Next, to train the model on ACDC, run the following:

python main.py

Testing

To validate the model performance, run the following:

python test.py

Citation

If you find this repository useful, please cite our work as follows:

@misc{basak2022embarrassingly,
      title={An Embarrassingly Simple Consistency Regularization Method for Semi-Supervised Medical Image Segmentation}, 
      author={Hritam Basak and Rajarshi Bhattacharya and Rukhshanda Hussain and Agniv Chatterjee},
      year={2022},
      eprint={2202.00677},
      archivePrefix={arXiv},
      primaryClass={eess.IV}
}

Acknowledgement

Thanks SSL4MIS for their wonderfurl work. Part of the code is borrowed from them. Please feel free to cite their work:

@misc{luo2020ssl4mis,
  title={SSL4MIS},
  author={Luo, Xiangde},
  year={2020}
}