Purpose: Magnetic resonance fingerprinting (MRF) unlocks incredible potential for efficient multi-parametric quantitative MRI. When coupled with MR-guided radiation therapy, MRF will push new boundaries for online functional plan adaptation. This work establishes the feasibility of MRF on a 0.35T MR-linac and introduces a novel deep learning-based method to improve quantitative parameter precision from clinically feasible MRF scan times.
Methods: MRF was benchmarked on a 0.35T MR-linac in the NIST/ISMRM System quantitative phantom. Gold standard multi-inversion recovery and multi-echo spin echo scans were acquired to derive reference quantitative T₁ and T₂ maps. A fully sampled MRF was designed and acquired in ~20 minutes. MRF-derived T₁ and T₂ times were compared to gold standard via linear regression and mean percent error. To assess the impact of accelerating MRF, the fully sampled dataset was undersampled to three radial k-space spokes/frame (28 seconds). Two acceleration approaches were tested: (1) uniformly spaced spoke angles with a linear reconstruction, and (2) spoke angles determined by a neural network with a deep learning regularized reconstruction. T₁ and T₂ times from both acceleration approaches were compared to fully sampled MRF using paired t-tests.
Results: Fully encoded MRF at 0.35T yielded precise T₁ and T₂ estimates relative to gold standard (mean percent errors of 4.65% and 8.24%, respectively) and correlation coefficients >0.996. Acceleration approach (1) produced T₁ and T₂ times different from those measured with the full 21-minute MRF scan (p=0.009 and p=0.013, respectively). The deep learning acceleration approach produced similar estimates for T₁ and T₂ (p=0.121 and p=0.363, respectively).
Conclusion: These encouraging results suggest that MR fingerprinting is feasible at 0.35T and, when coupled with deep learning-assisted acceleration, accurate MRF can be performed efficiently. Future work will optimize MRF for in vivo quantitative imaging, thereby opening the door for subvolume tumor targeting and treatment response assessment.
Funding Support, Disclosures, and Conflict of Interest: Support for this project was provided by the Bentson Foundation.
Not Applicable / None Entered.