Scale Attention Network (SA-Net) in Gossip Contrastive Mutual Learning (GCML)

SA-Net

SA-Net is the segmentation model we used for GCML study. It was evaluated on various medical image segmentation tasks under various imaging modalities with top performance, such as the 2nd place in BraTS 2021 challenge. More details can be found in the papers listed in the citation section.

Use cases

SA-Net on BraTS21 dataset

SA-Net on PanSeg dataset

SA-Net on HECKTOR dataset (coming soon)

GCML with SA-Net on BraTS21 dataset (coming soon)

GCML with SA-Net on PanSeg dataset (coming soon)

GCML with SA-Net on HECKTOR dataset (coming soon)

Source paper

https://doi.org/10.48550/arXiv.2503.03883

License

The code is licensed under GPL-3.0 license.

Citation

If you use this code for your research, please use the following BibTex entries:

@inbook{Yuan2022,
   author = {Yading Yuan},
   doi = {10.1007/978-3-031-09002-8_4},
   booktitle = {International MICCAI Brainlesion workshop},
   pages = {42-53},
   title = {Evaluating Scale Attention Network for Automatic Brain Tumor Segmentation with Large Multi-parametric MRI Database},
   year = {2022}
}
@article{Chen2025,
   author = {Jingyun Chen and Yading Yuan},
   doi = {10.1109/TMI.2025.3549292},
   issn = {0278-0062},
   journal = {IEEE Transactions on Medical Imaging},
   title = {Decentralized Personalization for Federated Medical Image Segmentation via Gossip Contrastive Mutual Learning},
   year = {2025}
}