tea321000 / WSL4MIS

Scribbles or Points-based weakly-supervised learning for medical image segmentation, a strong baseline, and tutorial for research and application.

Weakly-supervised learning for medical image segmentation (WSL4MIS).

• This project is developing, more details will be provided later, thanks for your attention. If you used this code in you research, please consider to cite the followings:

    @misc{wsl4mis2020,
      title={{WSL4MIS}},
      author={Luo, Xiangde},
      howpublished={\url{https://github.com/Luoxd1996/WSL4MIS}},
      year={2020}
    }
  

Dataset

• The ACDC dataset with mask annotations can be downloaded from: ACDC.
• The Scribble annotations of ACDC can be downloaded from: Scribble.
• The data processing code in Here or email me for pre-processed data.

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 ......

Follow official guidance to install Pytorch.

Usage

1. Clone the repo:

git clone https://github.com/Luoxd1996/WSL4MIS
cd WSL4MIS

2. Download and pre-process data and put the data in ../data/ACDC.

3. Train the model

cd code
python train_XXX_2D.py

4. Test the model

python test_2D_fully.py

5. Training curves on the fold1 Note: pCE means partially cross-entropy, TV means total variation, label denotes supervised by mask, scribble represents just supervised by scribbles.

Implemented methods

• pCE
• pCE + TV
• pCE + Entropy Minimization
• pCE + GatedCRFLoss

Acknowledgement

• The GatedCRFLoss is adapted from GatedCRFLoss for medical image segmentation.
• The codebase is adapted from our previous work SSL4MIS.