Conditional Diffusion-Generated Super-Resolution for Myocardial First-Pass Perfusion MRI

Motivation: This study addresses critical limitations in myocardial perfusion MRI, specifically restricted slice coverage and temporal resolution, which are essential for accurate perfusion assessment. Goal(s): To enhance myocardial perfusion MRI by achieving high slice coverage and temporal resolution, enabling real-time inline imaging display using low-resolution images for smoother workflow. Approach: We developed PerfGen, a conditional diffusion-based generative model that, combined with GRAPPA, achieves 6-9-fold acceleration by generating high-resolution images from low-resolution acquisitions. Results: Trained and validated on clinical rest and stress datasets, PerfGen reduced nRMSE by 20%, increased PSNR by 6%, and improved SSIM by 3%, complementing GRAPPA or other reconstruction methods. Impact: PerfGen enables high-quality myocardial perfusion MRI, complementing GRAPPA or similar methods to improve imaging speed with real-time displays. This conditional diffusion-based super-resolution approach generates multiple contrast phases and complements GRAPPA or similar methods, further advancing myocardial perfusion imaging.