Can On-Scanner Deep Learning Enable Clinical Viability of 0.55T for Knee MRI?

Motivation: 0.55T MRI is cost-friendly, environmentally-friendly, and patient-friendly, but can lack in image quality as compared with standard 3T images. AI-based image enhancement could resolve these differences, making 0.55T a clinically viable option for knee imaging. Goal(s): We examine whether 0.55T knee imaging could become a viable option for clinical practice with the help of a vision transformer-based image enhancer applied simultaneously during image acquisition. Approach: We collect a prospective set of knee images with and without enhancement on 0.55T as well as on 3T (20 total). We then test radiologist preference and clinically-integrated segmentation performance to examine the impact of image enhancement. Impact: Deep learning-based image enhancement applied automatically at scan time could make 0.55T a viable option for clinical knee imaging without disrupting workflow.