Integrating scout and guidance line-based retrospective motion correction into a 3D deep learning reconstruction for fast and robust brain MRI

Motivation: Rising medical imaging utilization and increasing use of automated support systems demand high-quality, fast, and reproducible/robust MRI techniques. Despite rapid scanning afforded by deep learning, motion remains a common source of artifacts. Goal(s): Integrate retrospective motion correction into a deep learning reconstruction to facilitate high-quality, fast, and motion-robust brain imaging. Approach: Scout and guidance line-based motion correction was implemented into MPRAGE, SPACE and SWI to enable rapid motion trajectory estimation. A data-consistency driven neural network reconstruction was adapted to perform network regularized motion correction. Results: Improved SNR and reduced motion artifacts are demonstrated in vivo using 4-6-fold accelerated scans with instructed subject motion. Impact: Retrospective motion correction was integrated into a deep learning reconstruction to facilitate fast and motion-robust 3D brain imaging across T1, T2, T2 FLAIR and T2*/SWI. This should add clinical value to routine brain exams and emerging neuro-degenerative screening protocols (ARIA).