Motivation: Ultrahigh-resolution MRI is very challenging due to long scan time and signal-to-noise trade-offs. Machine learning provides new opportunities but, to our knowledge, has not been demonstrated due to limited training data available, huge computational demands, and potential morphological distortions. Goal(s): To achieve generalizable ultrahigh-resolution MR brain imaging at 0.3 mm, using very limited data. Approach: We proposed a diffusion bridge with model-based fake feature correction using 0.3 mm priors from one brain image and 1.0 mm priors from 10,000 brain images. Results: Our approach successfully produced high-quality brain images at 0.3 mm, which were validated on both 13 public datasets and stroke patients. Impact: Conventional MRI scans of the brain are typically done at 1 mm resolution. Ultrahigh-resolution MRI will open up many opportunities for research and clinical applications. The proposed approach may also be useful for solving other imaging and processing problems.
Ke et al. (Tue,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: