Feasibility of Magnetic Resonance Fingerprinting on Aging MRI Hardware

Abstract

The purpose of this work is to evaluate the feasibility of performing magnetic resonance fingerprinting (MRF) on older and lower-performance MRI hardware as a means to bring advanced imaging to the aging MRI install base. Phantom and in vivo experiments were performed on a 1.5T Siemens Aera (installed 2015) and 1.5T Siemens Symphony (installed 2002). A 2D spiral MRF sequence for simultaneous T₁/T₂/M₀ mapping was implemented on both scanners with different gradient trajectories to accommodate system specifications. In phantom, for T₁/T₂ values in a physiologically relevant range (T₁: 195-1539 ms; T₂: 20-267 ms), scanners had strong correlation (R² > 0.999) with average absolute percent difference of 8.1% and 10.1%, respectively. Comparison of the two trajectories on the newer scanner showed differences of 2.6% (T₁) and 10.9% (T₂), suggesting a partial explanation of the observed inter-scanner bias. Inter-scanner agreement was better when the same trajectory was used, with differences of 6.0% (T₁) and 4.0% (T₂). Intra-scanner coefficient of variation (CV) of T₁ and T₂ estimates in phantom were <2.0% and in vivo were ≤3.5%. In vivo inter-scanner white matter CV was 4.8% (T₁) and 5.1% (T₂). White matter measurements on the aging scanner after two months were consistent, with differences of 1.9% (T₁) and 3.9% (T₂). In conclusion, MRF is feasible on an aging MRI scanner and required only changes to the gradient trajectory.

Keywords: accessible MRI; magnetic resonance fingerprinting; quantitative MRI; value MRI.

Figure 1

MRF sequence and k-space sampling trajectories. (a) For both scanners, the MRF sequence identically uses a variable repetition time (TR) and flip angle (FA) to induce sensitivity to T₁ and T₂ relaxation times. An inversion preparation pulse is applied prior to the first excitation. (b) Variable density spiral sampling trajectories are used on both scanners. The ‘weak gradient’ spiral designed for the old scanner MRF implementation (red) is more uniform than the ‘strong gradient’ spiral used for the new scanner MRF implementation (blue), resulting in a higher average acceleration factor (42 vs. 32, respectively).

Figure 2

MRF maps of the ISMRM/NIST quantitative imaging phantom from the new scanner with the standard MRF spiral trajectory (‘strong’ gradients), the new scanner with the modified spiral trajectory (‘weak’ gradients), and the old scanner with the modified (‘weak’) spiral trajectory. T₁ and T₂ maps show comparable quality without large differences in reference object relaxometry values. M₀ maps obtained from the two scanners have differing appearance potentially due to differences in receive coil sensitivities.

Figure 3

Correlation and repeatability analyses on the old scanner with the ‘weak gradient’ spiral and the new scanner with the ‘strong gradient’ spiral. (a,b) Correlation plots showing very strong inter−scanner correlation (p < 0.0001). Data points are averages from the five repeated acquisitions. (c,d) Bar plots of CV showing that the repeatability of T₁ and T₂ measurements (as computed from the five repeated scans) is comparable between the two scanners (<2% difference).

Figure 4

Bland–Altman analyses of T₁ and T₂ values obtained from the ISMRM/NIST MRI system phantom on the two scanners (old and new) showing patterns of bias and limits of agreement. (a,b) Comparison of the two evaluated spiral trajectories on the new scanner, the modified spiral with ‘weak’ gradients and the spiral conventionally used for MRF with ‘strong’ gradients. (c,d) Comparison of the new scanner with ‘weak gradient’ spiral versus the old scanner with the same ‘weak gradient’ spiral. (e,f) Comparison of the new scanner with ‘strong gradient’ spiral versus the old scanner with the ‘weak gradient’ spiral. (g,h) Comparison of T₁ and T₂ measurements on the old scanner across different days (test−retest assessment of reproducibility).

Figure 5

MRF maps of the brain from healthy subjects obtained from the new scanner (a) and the old scanner (b). T₁ and T₂ maps are of comparable overall quality, and values appear visually similar between corresponding tissue types. As also observed in phantom, M₀ maps differ in appearance between the two scanners, potentially due to differences in receive coil sensitivities.