TAT: Task-Adaptive Transformer for All-in-One Medical Image Restoration

PyTorch implementation for 《TAT: Task-Adaptive Transformer for All-in-One Medical Image Restoration》

🚀🚀🚀Check our paper collection of recent Awesome-Medical-Image-Restoration

Network Architecture

Dataset

You can download the preprocessed datasets for MRI super-resolution, CT denoising, and PET synthesis from Baidu Netdisk or Google Drive.

The original dataset for MRI super-resolution and CT denoising are as follows:

• MRI super-resolution: IXI dataset

• CT denoising: AAPM dataset

Visualization

You can use AMIDE to visualize the ".nii" file. Note that the color map for MRI and CT images is "black/white linear," while the color map for PET images is "white/black linear." Additionally, you need to rescale the PET image according to the voxel size specified in the paper.

Citation

If you find TAT useful in your research, please consider citing:

@inproceedings{yang2025tat,
  title={TAT: Task-Adaptive Transformer for All-in-One Medical Image Restoration},
  author={Yang, Zhiwen and Zhang, Jiaju and Yi, Yang and Liang, Jian and Wei, Bingzheng and Xu, Yan},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year={2025},
  organization={Springer}
}

Acknowledgments

The codebase is based on the awesome AMIR and Restore-RWKV repositories.