Diffusion Models with Awareness of Binary Signal Detection Tasks for Medical Image Denoising

Authors

Yaozhi Zhang, Wentao Chen, Weimin Zhou, and Yang Liu

Abstract

It is widely recognized that the evaluation and optimization of medical imaging system performance should be guided by task-based measures of image quality (IQ). We propose a task-aware diffusion model (TADM), a sampling-based method that incorporates channelized model observers, to denoise medical images for binary signal detection tasks. Our study employs Partial Least Squares (PLS) channels and uses singular value decomposition (SVD) to decompose the image into task-dense and task-sparse components. During the denoising process, TADMs simultaneously reduce noise in both task-dense and task-sparse components via corresponding posterior sampling. It is demonstrated that the proposed TADMs outperform Denoising Diffusion Restoration Models (DDRMs) in both physical IQ metrics (e.g. PSNR and FID) and task-based IQ metrics across a variety of image noise levels and datasets.

Article

BibTeX

@article{yaozhizhang2026,
  title={Diffusion Models with Awareness of Binary Signal Detection Tasks for Medical Image Denoising},
  author={Yaozhi Zhang and Wentao Chen and Weimin Zhou and Yang Liu},
  journal={Medical Imaging 2026: Image Perception, Observer Performance, and Technology Assessment},
  year={2026},
  month={}
}