Improved Resolution and Precision of whole-brain multi-parametric qMRI using Deep Learning Denoising and Super-Resolution Reconstruction

Motivation: High-resolution multi-parametric quantitative MRI (qMRI) is desirable for accurate tissue characterization. However, it comes with the cost of longer scan-time and reduced SNR. Parallel Imaging can mitigate the long scan-time, but it further compromises SNR. Goal(s): To use deep learning denoising and supper-resolution reconstruction (DLR) along with rapid k-space sampling methods to achieve a 5min 3D volumetric qMRI protocol. Approach: GRE images reconstructed with DLR were used for calculating quantitative maps and qualitative weightings. Quantitative values and image quality metrics were measured and compared between original and DLR. Results: DLR improved resolution, sharpness, SNR, CNR, and precision of qMRI while maintaining quantitative values. Impact: High-resolution and high-SNR whole-brain qMRI can be acquired within 5-6min using acceleration and DLR. qMRI's qualitative maps and quantitative weightings may be used for comprehensive evaluation of pathology and longitudinal follow-up.