Enabling Rapid Quantitative Multi-Tissue Parameter Mapping for Longitudinal Studies through Accelerated MRF with Deep Learning-Based Priors

Motivation: Quantitative MRI can improve tracking of subtle neurobiological changes in brain development and disease progression. Shorter scan times could expand its use, particularly for high-need, hard-to-scan groups like pediatric and older adults. Goal(s): To significantly reduce MRF scan times while maintaining high image quality and quantitative accuracy. Approach: We leverage prior MRF scans in reconstructing current scans using a tailored Dual-Stream-Attention-U-Net on the spatiotemporal subspace representation of MRF time-series data. Results: High-quality whole-brain T1 and T2 maps were obtained at 1mm-isotropic resolution from a 30s MRF scan in a longitudinal setting with the proposed approach, with good correspondence to the gold-standard 4-minute scan. Impact: This approach enables significantly shorter MRF scans in a longitudinal setting without compromising quality. It should facilitate more frequent monitoring of patients, particularly hard-to-scan groups like pediatric, and open avenues for further reducing scan times in quantitative imaging.