Machine Learning-Optimized Sampling Enables High-Resolution 3D MR Fingerprinting for Dynamic Quantification of Solute Transport in CSF

Motivation: Enhance spatiotemporal resolution of MR fingerprinting to dynamically track solute transport in cerebrospinal fluid. Goal(s): Increase undersampling capacity of MR fingerprinting by optimizing the sampling patterns for a stack-of-spiral trajectory. Approach: We developed a machine learning-based method to optimize rotation angles for spiral arms across kz partitions and time frames in 3D MRF acquisition. Optimized sampling pattern were tested in simulation studies and in vivo experiments. Results: An additional 2-fold undersampling was achieved, enabling simultaneous mapping of T1 and T2 across the whole mouse brain with 200-μm isotropic resolution in 4.3 min, allowing dynamically tracking contrast agents in CSF while maintaining anatomical detail. Impact: We present a novel method to leverage machine learning to improve sampling for MR fingerprinting. This method achieves simultaneous mapping of T1 and T2 in the whole mouse brain at 200-μm isotropic resolution in 4.3 min.