Introduction

One of the main reasons for using ultra-high fields (UHF) in proton magnetic resonance spectroscopy (1H-MRS) is for the gain in spectral resolution. However, this feature can only be exploited if the main magnetic field is sufficiently homogeneous.¹ In order to achieve this goal, effective approaches for B₀ shimming are required. Projection-based, B₀ shim techniques² have been proposed for fast, automatic 2^nd order B₀ shimming and are commonly used for single-voxel MRS.³ However, this technique is typically restricted to the adjustment of up to the 2^nd order shim coils and only provides shim values for one specific voxel position per shimming process. In contrast, fieldmap-based approaches⁴ can also include higher-order shim coils. Moreover, the acquired fieldmap can be used to calculate shim settings for multiple volumes, such as individual voxels and the entire brain. Here we investigate the advantage of this approach in comparison to projection-based techniques in terms of the achieved B₀ homogeneity and the required shimming time. Measurements were performed in three brain regions which are particularly problematic regarding B₀ inhomogeneity.

Methods

Experiments were performed in a 7T Siemens Terra scanner (Siemens, Erlangen, Germany), equipped with a 1Tx/32Rx head-coil and a full 2^nd- and partial 3^rd- order shim system. A T1-weighted mp2rage sequence⁵ was used to ensure correct positioning of the MRS voxels. Protocol parameters were: echo-time, TE=2.88ms; repetition-time, TR=4.5s; matrix-size=160×188×188 and voxel-size=1.2³mm³.
MRS spectra were acquired from three brain regions in three participants: striatum, left hippocampus and pons. MRS spectra were measured using a stimulated echo acquisition mode (STEAM).^6,7 For the hippocampus (voxel-size=27×13×10mm³), and pons (voxel-size=20×20×20mm³), the following protocol parameters were used: TE=4ms; mixing-time, TM=28ms; TR=8200ms; for the striatum (voxel-size=32×12×17mm³) TE/TM/TR=4/28/8000ms. Other parameters were: 50 averages; receive bandwidth, 6000Hz; vector-size, 2048. The radiofrequency power for the excitation flip-angle was calibrated for each subject. Water suppression was achieved using VAPOR⁸ interleaved with outer volume suppression pulses.⁶ One extra complete phase cycle was measured without the water suppression RF pulse to record a water peak reference for eddy-current corrections and absolute metabolite concentration calibrations.
Experiments were divided into two parts. In the first part, B₀ shim was performed using FASTESTMAP,⁹ with 1 linear and 3 non-linear iterations. The initial shim values were the ones obtained from one iteration of the shim technique provided by the vendor. Then, calibration of the excitation flip-angle was performed, and finally, the water-suppressed and water reference spectra were acquired. In the second part, the shim values were re-set to the default settings, a fieldmap was acquired, and a custom-written fieldmap-based shim technique was used to shim the MRS voxel. Subsequently, the water-suppressed and water reference spectra were re-measured. Finally, a whole-brain shim setting was calculated from the fieldmap and applied in order to test the feasibility of deriving multiple-shim settings from a single initial fieldmap acquisition. All spectra were quantified using LCModel with the basis of metabolites created using FID-A.¹⁰
The performance of each shim technique was evaluated in two ways. Firstly, we compared the full-width half maximum (FWHM) of the unsuppressed water peak via fitting with a Vought line shape.¹⁰ Secondly, we compared the Cramer-Rao lower bound (CRLB) of each metabolite.

Results and discussions

Fig.1 shows a schematic workflow of the experiments. Fig.2 shows the unsuppressed water peaks from different brain regions and subjects. We observe that the fieldmap-based shimming approach shows a narrower linewidth compared to FASTESTMAP. In the case of the fieldmap-based approach, FWHM of the hippocampus, striatum and pons were 15.4Hz, 18.4Hz and 19.7Hz, respectively, and for FASTESTMAP they were 22.1Hz, 20.4Hz and 27.7Hz, respectively. The bar plots in Fig.3a show the CRLB values for each of the detected metabolites. One can see that the CRLB values are, in general, lower in the case of the fieldmap-based shimming approach. Fig.3b shows the metabolite concentrations (in institute units, i.u.). In the case of the hippocampus, alanine and glycine were not reliably detected when FASTESTMAP was used. However, the better shimming achieved with the fieldmap-based approach enabled a reliable estimation. A similar situation was observed for NAAG and scyllo-inositol at the pons.
The SNR values, defined as the ratio of the maximum in the spectrum minus the baseline over the analysis window to twice the root-mean squares of the residuals, were also compared. The SNR values from the field-based approach for the three areas (15/24/22) were higher or equal to the SNR values from the FASTESTMAP approach (12/24/17). Besides the improved spectral quality for MRS acquisitions, the fieldmap-based shim algorithm also yielded high-quality shim settings for whole-brain acquisitions without the need for a further fieldmap acquisition. Thus, no additional acquisition time-penalty is introduced.

Conclusions

We showed the preliminary results of a study in which we aim to compare the projection-based shimming approaches with fieldmap-based approaches for single-voxel MRS studies at 7T. Our results demonstrate that the water linewidth and the water suppression were superior for the fieldmap-based shimming approach, as compared to the projection-based alternative. Moreover, the CRLB values were, in general, lower for the fieldmap-based approach. In the future, we plan to measure a larger number of subjects in order to improve the statistics of the results.

Acknowledgements

Supported by Shota Rustaveli National Science Foundation, Tbilisi, Georgia. We would like to acknowledge E.J. Auerbach and M. Marjanska (Center for Magnetic Resonance Research and Department of Radiology, University of Minnesota, USA) for the development of the STEAM sequence for the Siemens platform, which were provided by the University of Minnesota under a C2P agreement.