Optimization of variable-density undersampling for accelerated myocardial delayed enhancement imaging with deep learning reconstruction

Motivation: Myocardial delayed enhanced (MDE) imaging is the gold standard for assessing myocardial viability in various cardiac pathologies. However, long breath-hold is needed for MDE to achieve reasonable spatial resolution, hampering its utility for patients with insufficient breath-hold capability. Goal(s): The goal for this study is to optimize a variable-density undersampling pattern to achieve highly accelerated MDE imaging combined with deep learning reconstruction. Approach: The optimization was conducted with phantom and post-contrast in vivo studies. Results: The optimized undersampling pattern and deep learning reconstruction enable 4-time acceleration for phase-sensitive MDE imaging with comparable image quality to the reference image. Impact: The optimized variable-density undersampling pattern combined with deep learning reconstruction can potentially expand the clinical utility of MDE imaging to especially patient with insufficient breath-hold capability, and improve the patient comfort.