Attenuation correction for PET imaging using conditional denoising diffusion probabilistic model

During PET acquisition, attenuation correction is an indispensable procedure in which computed tomography (CT) image is usually regarded as the standard. However, attenuation correction in PET/MR systems is challenging because magnetic resonance (MR) images lack information on the attenuation coefficients. To address this issue, we propose a novel method to synthesize attenuation coefficients (CT) images from MR images using a conditional diffusion probabilistic model (con-DDPM). To enhance the constraint on generated images based on MR images, we introduce a condition encoder into the original diffusion model to extract features from MR images. And then, corresponding low-level features from the condition encoder and diffusion encoder are fused together. Results demonstrate that our proposed method can achieve more accurate attenuation coefficient synthesis than UNet and Generative Adversarial Network (GAN) models.