Volumetric Measurement Comparisons for Deep Learning Improved Ultra-Low Field MRI

Motivation: Accurate brain volumetrics from ultra-low-field MRI remains challenging, yet reliable brain region segmentation and volume estimates would be valuable in point-of-care settings and for monitoring neurological conditions in underserved areas. This study evaluates deep-learning models to expand ultra-low-field MRI's clinical and research applications where high-field MRI is unfeasible. Goal(s): To evaluate if deep-learning models can enhance ultra-low-field MRI images to match high-field MRI volumetric accuracy. Approach: Applied SynthSR and LoHiResGAN models to ultra-low-field 64mT MRI scans from 92-participants, comparing volumetric measurements of 19-brain regions to those from 3T MRI. Results: Enhanced images showed significantly improved volumetric accuracy, closely aligning with high-field MRI measurements. Impact: SynthSR and LoHiResGAN were evaluated for their unique approaches to enhancing ultra-low-field MRI images. Their ability to improve volumetric accuracy highlights their potential to support and expand access to high-quality neuroimaging in settings with limited access to high-field MRI.