Pre- to post-contrast medical image synthesis with outline-guide accelerate diffusion model

Contrast agents are widely used in medical imaging, which enhance the contrast of lesion region and promote detection and treatment. Due to contrast agents injected into the human body may cause adverse reactions and potential damage to sensitive organs. Then, some generative models attempt to synthesize post-contrast images from pre-contrast images to avoid the use of contrast agents. Within it, diffusion model has attracted significant interest due to its ability to effectively prevent mode collapse and ensure synthesis with clear details. However, existing diffusion models are suffering from easy-to-deform anatomical structure, indiscriminable noise- and edge-feature, and time-consuming iterations. To overcome the above problems, we, in this paper, propose a novel outline-guide accelerate diffusion model for pre- to post-contrast medical image synthesis. Specifically, we firstly use outline information from pre-contrast image as condition guidance to ensure anatomical structure consistency. Secondly, we design multi-frequency enhanced attention module to enable model to effectively distinguish between random noise and key features during feature extraction. Besides, we introduce non-uniform sampling strategy to reduce iterations to reduce training time and accelerate synthesis. Experiments show that ours maintains clearer detail texture and achieves high-quality post-contrast medical image synthesis with less training time. It achieves average increments of 10.2 % SSIM, 43.7 % PSNR, 9.5 % MSIM and decrements of 58.4 % NRMSE.