Super-Resolution Cardiac MR Tagging Using a Cascaded Conditional Diffusion-Based Generative Model

Motivation: This study aims to leverage cascaded diffusion-based conditional generative model to generate high-resolution MR tagging images from low-resolution acquisitions. Goal(s): To achieve 10-fold accelerated MR tagging with high-quality tag grids within 3 heartbeats. Approach: TagGen generates high-resolution MR tagging images from low-resolution acquisitions by incorporating parallel imaging and k-space consistency. Trained on retrospective and validated on prospectively acquired data, TagGen demonstrated superior tag grid fidelity, SNR, and overall image quality compared to existing approaches. Results: Preliminary studies show the model generalization across field strengths, imaging planes, and tag line spacings, highlighting its potential for highly accelerated MR tagging with high diagnostic quality. Impact: TagGen enables 10-fold accelerated, high-quality cardiac MR tagging, making accurate myocardial deformation assessment feasible within clinical workflows with greatly reduced scan time. This advancement broadens MR tagging accessibility and may enhance the detection of subtle heart motion abnormalities.