Accelerating 3D High-Resolution Brain MR Fingerprinting at 0.55T: Balanced Free-Precession meets Deep Learning

Motivation: 0.55T whole-body MR scanners may increase MRI accessibility. While lower SNR can lead to poor image quality, MRF with balanced gradients may leverage high SNR and benefit from lower B0 inhomogeneities. Goal(s): to generate fast 3D high-resolution MRF brain acquisition at 0.55T with improved SNR. Approach: Two 3D MRF acquisitions are implemented with balanced (bSSFP) and gradient-spoiled (FISP) schemes. Low-Rank is compared to a Deep-Image-Prior reconstruction, which embeds iterative intrinsic denoising. Scans were performed on phantoms and four healthy subjects. Results: The DIP-bSSFP MRF framework results in good accuracy, small coefficients of variation and yields good noise suppression and delineation of fine structures. Impact: MRF-bSSFP with DIP reconstruction enables fast high-resolution 3D quantitative T1 and T2 mapping at 0.55T in 3:50min, thus potentially enabling fast synthetic generation of contrast images with good noise suppression and delineation of fine structures.