MASSL-segmentation-framework

Multi-task Attention-based Semi-supervised Learning framework for image segmentation based on the paper published at MICCAI 2019 (https://arxiv.org/abs/1907.12303) by Shuai Chen, et al.

For questions please contact me through github or email directly.

Requirements

1. python 3
2. matplotlib
3. numpy
4. SimpleITK
5. sklearn
6. pytorch, torchvision
7. tqdm
8. skimage
9. scipy
10. elasticdeform

Training steps

1. Prepare 3D data for training, validation, and testing. Set the image patch size in module/common_module.py [BraTSshape]. Set folder path, preprocessing, and save as .npy files in BraTS2018_preprocess.py.
2. Set dataloader for pytorch, data split, and data augmentation in dataloader/BraTS18_dataloader.py.
3. Set random data seed, job you want to run, and data split you want to test in Sequance_BraTS18_epoch.py [for CNN baseline, pretraining methods, and MSSL method], or Sequance_BraTS18_epoch_attention.py [for MASSL method].
4. Run Sequance_BraTS18_epoch.py or Sequance_BraTS18_epoch_attention.py for training.

Testing

Change variable [Test_only=True] in Sequance_BraTS18_epoch.py or Sequance_BraTS18_epoch_attention.py and run again.

Citation

If you find the method useful for your research, please consider citing the paper:

@inproceedings{chen2019multi, title={Multi-Task Attention-Based Semi-Supervised Learning for Medical Image Segmentation}, author={Chen, Shuai and Bortsova, Gerda and Ju{'a}rez, Antonio Garc{'\i}a-Uceda and van Tulder, Gijs and de Bruijne, Marleen}, booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention}, pages={457--465}, year={2019}, organization={Springer} }

Poster: