Map-based B0 shimming for single voxel brain spectroscopy at 7T

Abstract

While B₀ shimming is an important requirement for in vivo brain spectroscopy, for single voxel spectroscopy (SVS), the role for advanced shim methods has been questioned. Specifically, with the small spatial dimensions of the voxel, the extent to which inhomogeneities higher than second order exist and the ability of higher order shims to correct them is controversial. To assess this, we acquired SVS from two loci of neurophysiological interest, the rostral prefrontal cortex (rPFC; 8 cc) and hippocampus (Hc; 9 cc). The rPFC voxel was placed using SUsceptibility Managed Optimization (SUMO) and an initial B₀ map that covers the entire cerebrum to cerebellum. In each location, we compared map-based shimming (Bolero) with projection-based shimming (FAST(EST)MAP). We also compared vendor-provided spherical harmonic first- and second-order shims with additional third- and fourth-order shim hardware. The 7T SVS acquisition used stimulated echo acquisition mode (STEAM) TR/TM/TE of 6 s/20 ms/8 ms, a tissue water acquisition for concentration reference, and LCModel for spectral analysis. In the rPFC (n = 7 subjects), Bolero shimming with first- and second-order shims reduced the residual inhomogeneity ${\sigma }_{B0}$ from 9.8 ± 4.5 Hz with FAST(EST)MAP to 6.5 ± 2.0 Hz. The addition of third- and fourth-order shims further reduced ${\sigma }_{B0}$ to 4.0 ± 0.8 Hz. In the Hc (n = 7 subjects), FAST(EST)MAP, Bolero with first- and second-order shims, and Bolero with first- to fourth-order shims achieved ${\sigma }_{B0}$ values of 8.6 ± 1.9, 5.6 ± 1.0, and 4.6 ± 0.9 Hz, respectively. The spectral linewidth, ${\Delta v}_{\sigma B0}$ , was estimated with a Voigt lineshape using ${\sigma }_{B0}$ and T2 = 130 ms. ${\Delta v}_{\sigma B0}$ significantly correlated with the Cramer-Rao lower bounds and concentrations of several metabolites, including glutamate and glutamine in the rPFC. In both loci, if the B₀ distribution is well described by a Gaussian model, the variance of the metabolite concentrations is reduced, consistent with the LCModel fit based on a unimodal lineshape. Overall, the use of the high order and map-based B₀ shim methods improved the accuracy and consistency of spectroscopic data.

Keywords: 7T; CRLB; linewidth; shimming; spectroscopy; spherical harmonic.

FIGURE 1

(A) MP2RAGE images showing the positioning of the location for the rostral prefrontal cortex voxel using SUsceptibility Managed Optimization (SUMO). The red box indicates the initial target location; the yellow box is the acceptable target region; and the green box is the SUMO-determined optimum position. The optimum position is selected based on a user-determined threshold, ${\sigma }_{\text{B0}}<\text{5}\text{Hz}$. In this example, the measured ${\sigma }_{\text{B0}}$ and the predicted ${\sigma }_{\text{B0}}$ after first- to fourth-order shimming at the initial location were 38.4 and 7.1 Hz, respectively. SUMO identified a location 4.24 mm away (3 mm in the superior and posterior directions) from the initial position with a predicted value of ${\sigma }_{\text{B0}}=4.2\text{Hz}$. The measured ${\sigma }_{\text{B0}}$ before and after first- to fourth-order shimming at this location was 34.5 and 3.9 Hz, respectively. (B) Positioning of the location for the hippocampal voxel along the temporal plane is determined by its well-known landmarks. MP2RAGE, magnetization-prepared 2 rapid gradient echo; ${\sigma }_{\text{B0}}$, standard deviation of the measured B₀ map distribution from the selected voxel.

FIGURE 2

(A–C) rPFC B₀ maps from a single subject after shimming with (A) Bolero first and second, (B) FAST(EST)MAP first and second, and (C) Bolero first to fourth. (D and E) Hc B₀ maps from a single subject (D) FAST(EST)MAP first and second, and (E) Bolero first to fourth. Each panel (A–E) shows the axial, sagittal, and coronal views of the B₀ maps (scale ±100 Hz) with magnified (panels A’–E’, scale ±25 Hz) views of the same data. The color scales are shown on the right. Hc, hippocampus; rPFC, rostral prefrontal cortex.

FIGURE 3

Histogram plots of the residual B₀ field after shimming from voxels obtained with different shim conditions from the (A) rPFC and (B) Hc. The data were acquired in four separate sessions (#1–#4). Bolero and FAST(EST)MAP data from the same session were acquired from the same location. For each plot, the ${\sigma }_{\text{B0}}$ and ${\sigma }_{\text{G}}$ are reported; the dashed black lines show the Gaussian lineshape with the indicated value of ${\sigma }_{\text{G}}$ fitted to the histogram data. Hc, hippocampus; rPFC, rostral prefrontal cortex; ${\sigma }_{\text{B0}}$, standard deviation of the measured B₀ map distribution from the selected voxel (in Hz); ${\sigma }_{\text{G}}$, center bandwidth of a Gaussian model of the measured B₀ map distribution (in Hz).

FIGURE 4

(A–C) Fitted spectra are shown from six paired datasets. With each column (A)–(C) of shim condition and locus, two different subjects are shown, with T1 MP2RAGE axial and sagittal images identifying the locus (red box) of acquisition. (A) rPFC acquired with Bolero first to fourth and FM first and second, (B) rPFC acquired with Bolero first and second and FM first and second, and (C) Hc spectra acquired with Bolero first to fourth and FM first and second. There are differences between the rPFC and Hc spectra; two of these differences are indicated with arrows at 3.75 ppm (summed amino acids) and choline (3.2 ppm). FM, FAST(EST)MAP; Hc, hippocampus; MP2RAGE, magnetization-prepared 2 rapid gradient echo; rPFC, rostral prefrontal cortex.

FIGURE 5

Correlations of $\text{Δ}{v}_{\text{LCM}}$ with calculated Voigt linewidths (A) $\text{Δ}{v}_{\sigma \text{G}}$ and (B) $\text{Δ}{v}_{\sigma \text{B0}}$. The regressions shown were taken without any offset, that is, $y=kx$ (both $x$ and $y$ parameters are measures of linewidth) and were all significant. The diagonal (black solid line) indicates the line of equality ($\text{Δ}{v}_{\text{LCM}}$ equals $\text{Δ}{v}_{\sigma \text{G}}$ or $\text{Δ}{v}_{\sigma \text{B0}}$). These data show that $\text{Δ}{v}_{\text{LCM}}$ is well matched to $\text{Δ}{v}_{\sigma \text{G}}$ for most of the data, while $\text{Δ}{v}_{\text{LCM}}$ matches $\text{Δ}{v}_{\sigma \text{B0}}$ primarily where $\text{Δ}{v}_{\sigma \text{B0}}$ is small, and where ${\sigma }_{\text{B0}}$ is small. The black circled markers of corresponding color identify the Bolero first to fourth shimmed data. Hc, hippocampus; RMSE, root-mean-square error; rPFC, rostral prefrontal cortex; SE, standard error; $\text{Δ}{v}_{\text{LCM}}$, LCModel-estimated spectral linewidth (in Hz); $\text{Δ}{v}_{\sigma \text{B0}}$, Voigt linewidth estimated by using ${\sigma }_{\text{B0}}$ (in Hz); $\text{Δ}{v}_{\sigma \text{G}}$, Voigt linewidth estimated by using ${\sigma }_{\text{G}}$ (in Hz); ${\sigma }_{\text{B0}}$, standard deviation of the measured B₀ map distribution from the selected voxel (in Hz); ${\sigma }_{\text{G}}$, center bandwidth of a Gaussian model of the measured B₀ map distribution (in Hz).

FIGURE 6

Regressions between LCM and shim parameters. (A and B) rPFC relationships between $\text{Δ}{v}_{\text{LCM}}$ with (A) CRLB, and (B) CSF-corrected tissue concentrations for Ins, Gln, and Glu. (C and D) rPFC correlations between $\text{Δ}{v}_{\sigma \text{B0}}$ with (C) CRLB, and (D) CSF-corrected tissue concentrations. (E) Hc correlations between $\text{Δ}{v}_{\sigma \text{B0}}$ with CRLB. Other Hc regressions for individual metabolites were not significant and thus are not shown. (F and G) rPFC and Hc correlations between the total scaled signal and (F) $\text{Δ}{v}_{\sigma \text{B0}}$, and (G) $\text{Δ}{v}_{\text{LCM}}$. For all panels, the significant regressions are identified with their $R^2$ values and linear equations. If the data are not significantly correlated, the regression line is not shown. In (B) and (D), the regression with Gln shows that a narrower linewidth correlates with less Gln concentration. Note that for (E), with the limited CRLB values (integer %), the Bolero first to fourth group data for Glu %SD CRLB values substantially overlap the Ins %SD CRLB values and are not visible. The black circled markers of corresponding color identify Bolero first to fourth shimmed data. CRLB, Cramer–Rao lower bound; CSF, cerebrospinal fluid; Gln, glutamine; Glu, glutamate; Hc, hippocampus; Ins, myo-inositol; LCM, LCModel, linear combination model software used for spectral fitting; rPFC, rostral prefrontal cortex; $\text{Δ}{v}_{\text{LCM}}$, LCModel-estimated spectral linewidth; $\text{Δ}{v}_{\sigma \text{B0}}$, Voigt linewidth estimated by using ${\sigma }_{\text{B0}}$; ${\sigma }_{\text{B0}}$, standard deviation of the measured B₀ map distribution from the selected voxel.

FIGURE 7

(A and B) Histograms (1.5 Hz bins) of the measured B₀ field across an rPFC voxel superimposed with the LCModel-fitted lineshape (solid line) after Bolero first- to fourth-order and FAST(EST)MAP shimming from matched voxels. The maximum value of the LCModel lineshape was centered on and scaled to be equal to the maximum value of the histogram. (C) The summed total signal (all metabolite signal scaled for proton multiplicity, macromolecule, and lipid content) ${\text{S}}_{\text{rPFC}}$ and ${\text{S}}_{\text{Hc}}$ are plotted against $f_{\text{G}}$. The variances for total signal in both the rPFC and Hc increase as $f_{\text{G}}$ declines. With a dividing threshold value for $f_{\text{G}}=2.0$, the mean and standard deviations for $f_{\text{G}}$ and total signal for the separated groups are shown with the cross hairs. The thick cross hairs are for the rPFC, and thin cross hairs are for the Hc. From this, F-tests show that the variances in the ${\text{S}}_{\text{rPFC}}$ and ${\text{S}}_{\text{Hc}}$ are significantly different between $f_{\text{G}}<2.0$ versus $f_{\text{G}}>2.0$: in the rPFC, ${\sigma }_{\text{SrPFC}}^2$ is 57.07 versus 949.28 (p < 0.05), and in the Hc, ${\sigma }_{\text{SHc}}^2$ is 145.35 versus 541.78 (p < 0.05). For both loci, the black circled markers of corresponding color identify the Bolero first to fourth shimmed data. (D) Plots of tissue concentrations for Ins, Gln, Glu, and GSH against the calculated $f_{\text{G}}$ parameter from the rPFC. Not all metabolites show an increase in variance as $f_{\text{G}}$ declines, although there is a trend to this effect consistent with the summed total signal data. (E) To allow a broader lineshape in the analysis, RFWHM = 5 was used to fit the data. The ${\text{S}}_{\text{rPFC}}$ and ${\text{S}}_{\text{Hc}}$ values are plotted against $f_{\text{G}}$. Gln, glutamine; Glu, glutamate; GSH, glutathione; Hc, hippocampus; Ins, myo-inositol; RFWHM, range for full width half maximum; rPFC, rostral prefrontal cortex.