Some important required packages include:
- Pytorch version >= 1.8.0.
- TensorboardX
- Python == 3.8
- Some basic python packages such as Numpy.
Follow official guidance to install Pytorch.
1. Contrastive Semi-supervised Learning for Domain Adaptive Segmentation Across similar Anatomical Structures
This repository provides the official code for "Contrastive Semi-supervised Learning for Domain Adaptive Segmentation Across similar Anatomical Structures".
- First, you should download the retinal dataset at REFUGE Challenge. We only used the 360 non-glaucoma images in this dataset and central-cropped and resized the images to 256. Second, you should download the CMR dataset at MS-CMRSeg 2019. We only used the bSSFP-sequence dataset and splited the 3D volumes into slice, and also central cropped and resized to 256.
- To train CS-CADA in circular structure segmentation, run:
python train_cscada.py
- To evaluate the trained model in CMR images for Left Ventricle (LV) and left ventricular Myocardium (Myo) segmentaiton, run:
python test_cscada_mscmrseg.py
If this project is helpful for your research, please cite the following works:
@article{gu2022contrastive,
title={Contrastive Semi-supervised Learning for Domain Adaptive Segmentation Across Similar Anatomical Structures},
author={Gu, Ran and Zhang, Jingyang and Wang, Guotai and Lei, Wenhui and Song, Tao and Zhang, Xiaofan and Li, Kang and Zhang, Shaoting},
journal={IEEE Transactions on Medical Imaging},
volume={42},
number={1},
pages={245--256},
year={2023},
publisher={IEEE}
}
@inproceedings{zhang2021ss,
title={SS-CADA: A semi-supervised cross-anatomy domain adaptation for coronary artery segmentation},
author={Zhang, Jingyang and Gu, Ran and Wang, Guotai and Xie, Hongzhi and Gu, Lixu},
booktitle={2021 IEEE 18th International Symposium on Biomedical Imaging (ISBI)},
pages={1227--1231},
year={2021},
organization={IEEE}
}
Part of the code is revised from UA-MT.
2. CDDSA: Contrastive Domain Disentanglement and Style Augmentation for Generalizable Medical Image Segmentation
This repository provides the official code for "CDDSA: Contrastive Domain Disentanglement and Style Augmentation for Generalizable Medical Image Segmentation". We first released our train and test codes.
- First, you should download the fundus dataset collected by Shujun Wang et al at DoFE project. We referenced the dataloader from this project and followed their training settings.
- To train CDDSA in fundus image segmentation, run:
python train_cddsa.py
- To evaluate the trained model in other domain for Optic Cup (OC) and Disc (OD) segmentaiton, run:
python test_cddsa_fundus.py
If this project is helpful for your research, please cite the following works:
@article{gu2022cddsa,
title={CDDSA: Contrastive Domain Disentanglement and Style Augmentation for Generalizable Medical Image Segmentation},
author={Gu, Ran and Wang, Guotai and Lu, Jiangshan and Zhang, Jingyang and Lei, Wenhui and Chen, Yinan and Liao, Wenjun and Zhang, Shichuan and Li, Kang and Metaxas, Dimitris N and others},
journal={arXiv preprint arXiv:2211.12081},
year={2022}
}
@inproceedings{gu2021domain,
title={Domain Composition and Attention for Unseen-Domain Generalizable Medical Image Segmentation},
author={Gu, Ran and Zhang, Jingyang and Huang, Rui and Lei, Wenhui and Wang, Guotai and Zhang, Shaoting},
booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
pages={241--250},
year={2021},
organization={Springer}
}