[论文解读] Unsupervised Medical Image Translation with Adversarial Diffusion Models

Muzaffer Özbey, Onat Dalmaz+5 more|arXiv (Cornell University)|Jul 17, 2022

Generative Adversarial Networks and Image Synthesis被引用 32

一句话总结

SynDiff 引入了一种条件对抗扩散框架，用于无监督医学图像翻译，在使用未配对数据的情况下实现源模态与目标模态之间的高保真映射。它将快速的对抗扩散过程与一个循环一致、非扩散的模块耦合起来，以在如多对比度 MRI 与 MRI-CT 等模态之间进行翻译。

ABSTRACT

Imputation of missing images via source-to-target modality translation can improve diversity in medical imaging protocols. A pervasive approach for synthesizing target images involves one-shot mapping through generative adversarial networks (GAN). Yet, GAN models that implicitly characterize the image distribution can suffer from limited sample fidelity. Here, we propose a novel method based on adversarial diffusion modeling, SynDiff, for improved performance in medical image translation. To capture a direct correlate of the image distribution, SynDiff leverages a conditional diffusion process that progressively maps noise and source images onto the target image. For fast and accurate image sampling during inference, large diffusion steps are taken with adversarial projections in the reverse diffusion direction. To enable training on unpaired datasets, a cycle-consistent architecture is devised with coupled diffusive and non-diffusive modules that bilaterally translate between two modalities. Extensive assessments are reported on the utility of SynDiff against competing GAN and diffusion models in multi-contrast MRI and MRI-CT translation. Our demonstrations indicate that SynDiff offers quantitatively and qualitatively superior performance against competing baselines.

研究动机与目标

• 推动多模态成像，以改善诊断评估和协议多样性。
• 用数据驱动先验解决跨模态的病态、非线性目标合成问题。
• 在保持解剖结构一致性的前提下，实现来自未配对的源-目标数据集的无监督翻译。

提出的方法

• 提出 SynDiff，一种通过大步长 k 实现快速采样的条件扩散模型。
• 引入一个源条件的对抗投影器，在源图像的引导下执行去噪。
• 使用将扩散与非扩散模块相结合的循环一致性架构，以实现未配对训练。
• 使用非扩散模块从目标估计成对的源图像，从而实现无监督学习。
• 在两种模态上联合训练，结合循环一致性损失和对抗损失。

实验结果

研究问题

• RQ1 Can SynDiff achieve high-fidelity, realistic translations between modalities using unpaired data?
• RQ2 Does the conditioned diffusion with adversarial projection improve sampling efficiency and accuracy over standard diffusion models?
• RQ3 How does the proposed cycle-consistent architecture perform for unsupervised learning in multi-contrast MRI and MRI-CT translation?
• RQ4 How does SynDiff compare to GAN and diffusion baselines in terms of image quality and diversity?

主要发现

• SynDiff 在定量和定性方面均优于竞争基线。
• The approach enables effective translation across multi-contrast MRI and MRI-CT using unpaired data.
• A fast diffusion scheme with large step size and an adversarial projector maintains sampling fidelity.
• A cycle-consistent framework with coupled diffusive and non-diffusive modules enables unsupervised training.
• The method provides high-fidelity target images guided by anatomically corresponding source images.
• Code for SynDiff is publicly available.