Improved Reconstruction Speed for 5D Free Running using a Variable Projection Augmented Lagrangian (VPAL) Method with Targeted Descent Step

Motivation: 5D free-running whole heart CMR offers CT-quality images but requires hours-long reconstruction time, preventing clinical usage. Therefore, a more efficient reconstruction algorithm is needed. Goal(s): We propose to use the advanced numerical algorithm to reduce the reconstruction time while preserving image quality. Approach: A variable projection augmented Lagrangian (VPAL) method with targeted update step for 5D image reconstruction was developed and compared with the state-of-the-art alternating direction method of multipliers (ADMM) on phantom and 15 pediatric patients' datasets. Results: When compared to the ADMM method, VPAL reduced the reconstruction time by 70%, preserved image similarity, had equivalent ejection fraction measurements and radiologist ratings. Impact: This study shows that using an advanced numerical algorithm for highly under-sampled MR reconstruction both reduces computational time and results in better image quality for diagnostics, bringing 5D free-running imaging is closer to clinical usage.