Measurement-conditioned Denoising Diffusion Probabilistic Model for Under-sampled Medical Image Reconstruction
Yutong Xie Academy for Advanced Interdisciplinary Studies Peking University Beijing, 100871、Quanzheng Li MGH/BWH Center for Advanced Medical Computing and Analysis Gordon Center for Medical Imaging, Department of Radiology Massachusetts General Hospital and Harvard Medical School Boston, MA 02114
Abstract
We propose a novel and unified method, measurement-conditioned denoising diffusion probabilistic model (MC-DDPM), for under-sampled medical image reconstruction based on DDPM. Different from previous works, MC-DDPM is defined in measurement domain (e.g. k-space in MRI reconstruction) and conditioned on under-sampling mask. We apply this method to accelerate MRI reconstruction and the experimental results show excellent performance, outperforming full supervision baseline and the state-of-the-art score-based reconstruction method. Due to its generative nature, MC-DDPM can also quantify the uncertainty of reconstruction. Our code is available on github111https://github.com/Theodore-PKU/MC-DDPM.
中文速览
医学影像的欠采样重建（under-sampled reconstruction）长期面临一个矛盾：如何在保证数据一致性的同时，充分利用生成模型的表达能力。作者提出了一种名为 MC-DDPM（测量域条件去噪扩散概率模型）的新方法，核心创新在于把扩散过程从图像域搬到了测量域（如 MRI 的 k 空间），并以欠采样掩码作为条件，让数据一致性天然内嵌于模型之中，无需在采样时额外执行数据一致性投影步骤。在 fastMRI 膝关节数据集上的加速 MRI 重建实验表明，MC-DDPM 在 PSNR 和 SSIM 指标上超越了全监督基线 U-Net 和当时最先进的基于得分函数的重建方法。此外，由于该方法具有生成模型的本质，可通过多次采样用蒙特卡罗方式估计像素级不确定性，为临床诊断提供了额外的可靠性参考，具有重要的实用价值。
原文 arXiv:2203.03623;中英对照 + 大白话阅读 https://aha.fim.ai/paper/2203.03623v1