Rapid Magnetic Resonance Fingerprinting for Brain Glioma with Synergistic Excitation–Sampling Optimization: Retrospective Development and Prospective Evaluation
Abstract
Purpose
Magnetic Resonance Fingerprinting (MRF) provides quantitative T1/T2 mapping that can support diagnosis, treatment planning, and longitudinal assessment in brain glioma. Clinical time limits necessitate accelerated MRF, which may introduce artifacts and bias. We developed a method that synergistically optimizes excitation and sampling for faster, higher-quality brain MRF.
Methods
We performed synergistic optimization of sequence scheduling (flip angle (FA) and repetition time (TR)) and sampling spiral trajectories using a differentiable signal model. The model incorporated field inhomogeneities, coil sensitivities, and off-resonance. Reconstruction was fixed (adjoint NUFFT plus dictionary matching) to emphasize deployability. Each updated trajectory was projected to satisfy gradient amplitude, slew rate, and peripheral nerve stimulation (PNS) limits. Retrospective training used multi-coil brain MRF from 12 glioma patients and 6 healthy volunteers scanned at 3T; 2 patients and 2 volunteers were held out for internal testing. Low-rank total-variation maps reconstructed from 1,000 dynamics served as reference. External validation used five volunteers from a separate 3T vendor dataset. Designs were compared at 200 dynamics using PSNR, NMSE, and SSIM.
Results
Retrospectively, at 200 dynamics (~2.2 s/slice), the design optimized by our method delivered the highest fidelity versus eight baseline designs: PSNR 39.30 dB (internal) and 26.92 dB (external), with the lowest NMSE. Under added synthetic noise, median T1/T2 errors were minimal. In glioma patients, maps showed fewer streaks and reduced spatial blotching. Agreement improved within lesion and peritumoral ROIs alongside increased margin conspicuousness. Prospective feasibility was demonstrated in healthy volunteer brain scans; a NIST phantom reproduced expected compartmental T1/T2 ordering with sharper vial boundaries and improved intra-vial uniformity.
Conclusion
Synergistic excitation–sampling optimization reduces aliasing and bias at clinically short scan times, producing robust quantitative maps using standard reconstruction. This enables more reliable glioma evaluation for diagnosis, planning, and follow-up. It supports routine brain tumor quantitative MRI and vendor-neutral multicenter studies.