Clinical Validation of Motion-Informed 3D Deep Learning Reconstruction in Patients with Cognitive Impairment at 3T

Motivation: Advances in network-regularized motion correction and deep learning (DL) reconstruction techniques require clinical validation in motion-prone patients. Goal(s): To validate the performance of motion-informed 3D DL reconstruction. Approach: Scout accelerated motion correction was integrated into data-consistency driven DL reconstruction on MPRAGE. The method was validated on healthy volunteers performing instructed motion and patients undergoing clinical evaluation for memory loss. Validation included quantitative morphometry analysis and assessment by board-certified neuroradiologists. Results: Compared to reference standard reconstruction, the proposed approach demonstrated improved SNR, superior morphometric accuracy (validated against motion-free reference data) reduced visual motion artifacts, and improved overall image quality in clinical cases. Impact: Deep learning-based reconstruction with integrated retrospective motion correction produced high-quality 3D MPRAGE in patients with cognitive decline. With clinically feasible processing times and robust performance, this technique shows promise for a wide range of neuroimaging protocols.