Brain MRI Synthesis using Enhanced Denoising Diffusion Probabilistic Models from Brain PET and CT scans

Denoising diffusion probabilistic models (DDPM) has shown superior performance in image synthesis for various fields. In this study, we propose an enhanced DDPM framework with the joint probability for synthesizing brain MRI from brain PET/CT. Although PET/MRI scans provide high-resolution structural information for partial evaluation and image reconstruction, it is costly and time-consuming compared to PET/CT. Also, synthesizing MRI from PET/CT including anatomical details has not been extensively studied. We achieved significant performance of proposed network compared with U-Net and cGAN in various input.