Proton metabolic mapping of the brain at 7 T using a two-dimensional free induction decay-echo-planar spectroscopic imaging readout with lipid suppression

Abstract

The increased signal-to-noise ratio (SNR) and chemical shift dispersion at high magnetic fields (≥7 T) have enabled neuro-metabolic imaging at high spatial resolutions. To avoid very long acquisition times with conventional magnetic resonance spectroscopic imaging (MRSI) phase-encoding schemes, solutions such as pulse-acquire or free induction decay (FID) sequences with short repetition time and inner volume selection methods with acceleration (echo-planar spectroscopic imaging [EPSI]), have been proposed. With the inner volume selection methods, limited spatial coverage of the brain and long echo times may still impede clinical implementation. FID-MRSI sequences benefit from a short echo time and have a high SNR per time unit; however, contamination from strong extra-cranial lipid signals remains a problem that can hinder correct metabolite quantification. L2-regularization can be applied to remove lipid signals in cases with high spatial resolution and accurate prior knowledge. In this work, we developed an accelerated two-dimensional (2D) FID-MRSI sequence using an echo-planar readout and investigated the performance of lipid suppression by L2-regularization, an external crusher coil, and the combination of these two methods to compare the resulting spectral quality in three subjects. The reduction factor of lipid suppression using the crusher coil alone varies from 2 to 7 in the lipid region of the brain boundary. For the combination of the two methods, the average lipid area inside the brain was reduced by 2% to 38% compared with that of unsuppressed lipids, depending on the subject's region of interest. 2D FID-EPSI with external lipid crushing and L2-regularization provides high in-plane coverage and is suitable for investigating brain metabolite distributions at high fields.

Keywords: 7 T; EPSI; FID; L2-regularization; MRSI; crusher coil; lipid suppression; pulse acquire.

FIGURE 1

Pulse sequence diagram used for the free induction decay–echo‐planar spectroscopic imaging (FID‐EPSI) acquisition. The sequence consists of variable power and optimized relaxation delays (VAPOR) water suppression and the trigger signal generated by spectrometer initialization of an external amplifier that drives the crusher coil, followed by the FID‐EPSI readout at 7 T. The duration of the lipid suppression gradient is 1.7 ms. The FID‐EPSI sampled 512 spectral points with a 2379 Hz spectral bandwidth and a 185 ms readout time

FIGURE 2

Spectral ghosting procedure for the echo‐planar spectroscopic imaging (EPSI) data via reconstruction steps (see Table 1). Gradient‐echo image (B) showed using a calibration phantom A (a sphere of 10 cm diameter) placed in the isocenter of the MR bore. (A) The effect on the acquired k‐t signal of the free induction decay (FID)‐EPSI readout for the phase‐corrected data by the following steps at the center location in (B). The alignment of k‐t space (A) is shown using the water reference signal, and (C) FID signals are presented at locations I and II in (B)

FIGURE 3

(A) Low resolution of T₁‐weighted images with/without lipid suppression using a crusher coil. (B, C) Lipid contamination maps in a matrix of 38 × 38 voxels. The lipid maps are calculated by integrating the spectrum between 0.8 and 1.6 ppm. The lipid binary mask (light blue line) in the lipid contamination map without lipid suppression (B, top) is used to the map (B, bottom) with lipid suppression. The area of the residual lipids (red arrow) is indicated in the lipid mask. (C) Maps with a factor of 20 multiplied intensity scale visualize lipid signal leakage near the skull in more detail. A point‐spread function pattern is visible in the unsuppressed case (C, top). This pattern is reduced in the suppressed case, where the lipid suppression reduction factor was 7 in V1, and 2 in V2 and V3. (D) Distribution of the (un)suppressed lipid signal intensity of the voxels in the entire region of interest per volunteer

FIGURE 4

(A) Anatomical brain image with two voxel locations (top) and the three different regions of interest (ROIs) (bottom). (B) MR spectra obtained with crusher coil and L2‐regularization of the two voxels in (A) with LCModel fit. (C) Concentration ratios (tCho, Glx, and tNAA) to total Cr (tCr) as calculated by LCModel for each ROI. The number of voxels of these ROIs (R1/R2/R3) in a matrix size of 38 × 38 is 48/16/38 for volunteer 1 (V1), 52/25/26 for volunteer 2 (V2), and 34/29/27 for volunteer 3 (V3) (see Figure S2)

FIGURE 5

Quality assurance maps (signal‐to‐noise ratio [SNR] and full width at half maximum [FWHM]) and the reconstructed metabolite ratio (divided by creatine + phosphocreatine [tCr]) maps of N‐acetyl‐aspartate + N‐acetyl aspartate glutamate (tNAA), choline + glycerophosphorylcholine + phosphorylcholine (tCho), glutamine + glutamate (Glx), and myo‐inositol + glycine (mI + Gly) were generated using different lipid‐suppression strategies. In the unsuppressed lipid case, clear effects of the lipid point‐spread function were evident for tNAA, Glx, and mI + Gly. Optimal lipid suppression was achieved with the combination of the crusher coil (direct lipid suppression) and L2‐regularization (lipid removal). For visualization, T1‐weighted image and metabolite ratio maps were interpolated by a factor of 2 (the final matrix size is 76 × 76)

FIGURE 6

(A) T1‐weighted image with the delineation of three regions of interest (ROIs) at V1. Region 1 contains predominantly white matter, and region 2 contains predominantly gray matter + CSF, while region 3 highlights boundary effects related to the lipid suppression field of the crusher coil. (B) Voxel average (solid line) and standard deviation (shaded area) of the fitted MR spectra for each ROI and each experimental variant; (red) crusher coil and L2‐regularization, (green) crusher coil and no L2‐regularization, (purple) no crusher coil and L2‐regularization, and (black) no crusher coil and no L2‐regularization. (C) Individual fits of tNAA, Glx, tCr, tCho, and the baseline in the four experimental settings (red, green, purple, and black). Note that spectra have been shifted to facilitate visual comparison