Introduction

Inhomogeneous Magnetization Transfer (ihMT) imaging is a recent technique that has demonstrated good sensitivity to myelination^1,2. When performed at relatively low field (e.g. 1.5T) or in small field of view (e.g. in small animal applications) a rather homogeneous B₁⁺ environment is typically observed. However, special attention must be paid to the local B₁⁺ variations when applications at high field or on large samples are targeted, to prevent shadowing artifacts inherent to B₁⁺ inhomogeneities.

Mchinda et al. have evidenced different B₁⁺ dependency regimes of the ihMT ratio metric (ihMTR) as a function of the RF energy concentration of MT saturation pulses using a sensitivity-boosted ihMT gradient echo (ihMT-GRE) sequence at 1.5T³. It was shown that concentrating RF power (i.e. use of long TR) may yield lower relative variations of ihMTR with B₁⁺ variations (illustrated on Figure 1 with simulated ihMTR data), hence holding promise for high field applications. These findings have been confirmed experimentally in a recent preliminary study performed on brain at 3T⁴.

In this work, we expand on these results and present a thorough analysis of the ihMTR metric sensitivity to local B₁⁺ values in order to provide hints for the optimization of ihMT-GRE sequences at 3T for human brain applications.

Materials and methods

Experiments were performed on a 3T clinical scanner (Verio, Siemens Healthineers, Erlangen, Germany) with body coil transmission and a 32-channel receive head coil on healthy volunteers. Every protocol included a 3D anatomical sequence (MPRAGE) as well as a B₁⁺ mapping sequence (pre-saturated turbo FLASH). Two sets of 3D ihMT-GRE sequences (referred hereafter to as Configuration #1 and #2 with parameters listed in Table 1) were acquired on three volunteers each, with a constant nominal B_1,RMS of 2.7 µT. Both configurations mostly differ in their number of RF pulses per saturation burst³, with Configuration #2 allowing for stronger RF energy concentration (achieved with longer TR) considering the peak power limitations of the system. Each ihMT experiment was performed twice: 1) with a nominal B₁⁺ adjustment, and 2) with a B₁⁺ distribution deliberately attenuated by 20%, achieved by setting the transmitter reference voltage to 80% of its nominal value (respectively referred hereafter to as V_ref and V_ref,80%).

Analyses were performed on the whole white matter area segmented from the MPRAGE using FreeSurfer⁵, and projected over ihMTR and resampled B₁⁺ maps. All extracted voxels were pooled across the three subjects for each generated map. Resulting distributions were used for the computation of mean and standard deviation of ihMTR from the V_ref datasets. Composite maps of voxel-wise ihMTR relative variation between V_ref and V_ref,80% (δ_ihMTR) were used to compute the following statistics: i) mean and standard deviation of δ_ihMTR; ii) voxel fraction with a relative variation within a 10% error margin: δ_{ihMTR-[-10%;10%]}; iii) root-mean-square error with respect to the ideal case of a 0% ihMTR relative variation (perfect immunity to B₁⁺ variations): δ_ihMTR-RMS. Insight into the B₁⁺ spatial distribution was further provided by investigating the same metrics in clusters of WM voxels with specific B₁⁺ values ranging from 0.86 to 1.16 (1.00 referring to the nominal B₁⁺ value), with a class size of 0.02.

Results

Representative images and metrics pertaining to the whole WM area are reported in Figure 2. Overall the highest average ihMTR were obtained in both configurations for an intermediate TR of 145 ms. Regarding immunity to B₁⁺ variations the best scores (minimal bias: lowest δ_ihMTR and δ_ihMTR-RMS, highest δ_{ihMTR-[-10%;10%]}) were unambiguously obtained at TR=265 ms for Configuration #1. For Configuration #2, the best settings lie in the range 265-ms ≤ TR ≤ 345-ms depending on the considered figure of merit. Of interest, the B₁⁺-clustered data analysis, shown in Figure 3 for the absolute ihMTR measured at V_ref and on Figure 4 for δ_ihMTR, is generally consistent with the simulation behaviours from Figure 1: short TRs leading to quasi-linear increase of ihMTR with B₁⁺ in the tested range, and associated with a high B₁⁺-induced bias, and long TRs leading to a bell-shaped dependency of ihMTR with B₁⁺, associated with minimal B₁⁺-induced bias.

Discussion and conclusion

Sensitivity of ihMTR with B₁⁺ variations has been investigated experimentally and agrees well with theoretical expectations. From the simulations presented in Figure 1, it can indeed be qualitatively understood that RF energy concentration should reduce the sensitivity to B₁⁺. Noteworthy the bell-shaped curve obtained in the more concentrated case explains that δ_ihMTR may change sign (getting negative) in areas of strong B₁⁺, consistent with the cluster analysis displayed in Figure 4 (Configuration #2).

This study indicates that it is not possible to cancel the B₁⁺ induced bias in all voxels simultaneously with the proposed approach, suggesting that the fraction of voxels within acceptable error margins or the mean root-mean-square error should be better indicators of the overall performance of ihMTR than the more usual mean ihMTR relative variation. Noteworthy, the settings the most robust to B₁⁺ variations do not provide the highest average ihMTR values. A tradeoff is therefore to be made between high ihMTR and reduced sensitivity to B₁⁺ inhomogeneities.

Acknowledgements

This work was supported by the SATT Sud-Est (France), the French Association pour la Recherche sur la Sclérose En Plaques (ARSEP), Roche Research Foundation (Switzerland) and French National Research Agency, ANR [ANR‐17‐CE18‐0030]. This work was performed by a laboratory member of France Life Imaging network (grant ANR-11-INBS-0006).