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
- python 3
- matplotlib
- numpy
- SimpleITK
- sklearn
- pytorch, torchvision
- tqdm
- skimage
- scipy
- elasticdeform
Training steps
- 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.
- Set dataloader for pytorch, data split, and data augmentation in dataloader/BraTS18_dataloader.py.
- 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].
- 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:
