(IEEE TIP) Degradation-Aware Prompted Transformer for Unified Medical Image Restoration

Jinbao Wei, Gang Yang, Zhijie Wang, Shimin Tao, Aiping Liu, Xun Chen,

Introduction

Medical image restoration (MedIR) aims to recover high-quality images from degraded inputs, yet faces unique challenges from physics-driven degradations and multi-modal task interference. While existing all-in-one methods handle natural image degradations well, they struggle with medical scenarios due to limited degradation perception and suboptimal multi-task optimization. In response, we introduce DaPT, a Degradation-aware Prompted Transformer, which integrates dynamic prompt learning and modular expert mining for unified MedIR. First, DaPT introduces spatially compact prompts with optimal transport regularization, amplifying inter-prompt differences to capture diverse degradation patterns. Second, a mixture of experts dynamically routes inputs to specialized modules via prompt guidance, resolving task conflicts while reducing computational overhead. The synergy of prompt learning and expert mining further enables robust restoration across multi-modal medical data, offering a practical solution for clinical imaging. Extensive experiments across multiple modalities (MRI, CT, PET) and diverse degradations, covering both in-distribution and outof-distribution scenarios, demonstrate that DaPT consistently outperforms state-of-the-art methods and generalizes reliably to unseen settings, underscoring its robustness, effectiveness, and clinical practicality.

Citations

You may want to cite:

@article{wei2025degradation,
  title={Degradation-Aware Prompted Transformer for Unified Medical Image Restoration},
  author={Wei, Jinbao and Yang, Gang and Wang, Zhijie and Tao, Shimin and Liu, Aiping and Chen, Xun},
  journal={IEEE Transactions on Image Processing},
  year={2025},
  publisher={IEEE}
}

Acknowledgments

This code is built on AMIR. Thank the authors for sharing their awesome codes!