Knee Imaging at 0.05 T with Deep Learning Superresolution Reconstruction

Motivation: There has been significant development in ultralow-field (ULF) MRI for low-cost, shielding-free, and point-of-care extremity applications. However, its image quality remains poor, and scan times are long. Goal(s): We aim to advance the speed and quality of knee ULF MRI using 2D partial Fourier sampling and deep learning image formation. Approach: A fast acquisition and deep learning reconstruction framework to accelerate knee MRI at 0.05 tesla was proposed. Results: 3D deep learning leverages high-field knee anatomy data to enhance image quality, reduce artifacts and noise, and improve spatial resolution. Impact: The method effectively overcomes the low-signal barrier, reconstructing fine anatomical structures at 0.05 Tesla that are reproducible within subjects and consistent across two protocols. It enables rapid, high-quality ULF MRI for potential point-of-care applications.