Atlas-based GABA mapping with 3D MEGA-MRSI: Cross-correlation to single-voxel MRS

Abstract

The purpose of this work is to develop and validate a new atlas-based metabolite quantification pipeline for edited magnetic resonance spectroscopic imaging (MEGA-MRSI) that enables group comparisons of brain structure-specific GABA levels. By using brain structure masks segmented from high-resolution MPRAGE images and coregistering these to MEGA-LASER 3D MRSI data, an automated regional quantification of neurochemical levels is demonstrated for the example of the thalamus. Thalamic gamma-aminobutyric acid + coedited macromolecules (GABA+) levels from 21 healthy subjects scanned at 3 T were cross-validated both against a single-voxel MEGA-PRESS acquisition in the same subjects and same scan sessions, as well as alternative MRSI processing techniques (ROI approach, four-voxel approach) using Pearson correlation analysis. In addition, reproducibility was compared across the MRSI processing techniques in test-retest data from 14 subjects. The atlas-based approach showed a significant correlation with SV MEGA-PRESS (correlation coefficient r [GABA+] = 0.63, P < 0.0001). However, the actual values for GABA+, NAA, tCr, GABA+/tCr and tNAA/tCr obtained from the atlas-based approach showed an offset to SV MEGA-PRESS levels, likely due to the fact that on average the thalamus mask used for the atlas-based approach only occupied 30% of the SVS volume, ie, somewhat different anatomies were sampled. Furthermore, the new atlas-based approach showed highly reproducible GABA+/tCr values with a low median coefficient of variance of 6.3%. In conclusion, the atlas-based metabolite quantification approach enables a more brain structure-specific comparison of GABA+ and other neurochemical levels across populations, even when using an MRSI technique with only cm-level resolution. This approach was successfully cross-validated against the typically used SVS technique as well as other different MRSI analysis methods, indicating the robustness of this quantification approach.

Keywords: GABA; MEGA-LASER; MEGA-PRESS; MRSI; validation.

Fig. 1.

(a) Localization of a 30 x 25 x 25 mm SV ROI depicted on axial, coronal and sagittal 3D T1-weighted images in the thalamic region. (b) FOV (200 x 200 x 170 mm) and VOI placement for 3D MRSI. The yellow box represents the FOV; the red box displays the VOI; green grids indicate the spatial acquisition matrix.

Fig. 2.

(a) Example of MRSI VOI (brown), thalamus mask (red) and SV ROI mask (blue) overlaid on anatomical images; (b) representative GABA+ map overlaid on anatomical images in axial, sagittal and coronal planes; (c) an example of 4 manually picked voxels shown in the turquoise box (4VX approach) overlaid on the GABA+ map and anatomical images to quantify thalamic neurochemical levels

Fig. 3.

Post processing pipeline for calculating neurochemical levels and their corresponding ratios to tCr, and comparison of the measurements from SV MEGA-PRESS and MEGA-LASER 3D MRSI.

Fig. 4.

(a) Representative difference spectrum (left) and edit OFF spectrum (right) acquired with SV MEGA-PRESS; (b) representative difference spectra from the VOI within one axial slice; (c) examples from seven subjects showing the overlay of the averaged spectra from the MEGA-LASER 3D MRSI voxels within the thalamus (black) and spectra acquired from SV MEGA-PRESS (red), with negative NAA and baseline aligned.

Fig. 5.

Bland-Altman plots for GABA+/tCr, tNAA/tCr and Glx/tCr from the SVS measurement and the three MRSI quantification approaches (ATL, ROI and 4VX), showing the mean difference (MD, red solid lines) and 95% confidence intervals of limits of agreement (red dotted lines).

Fig. 6.

Scatter plots with lines of regression through the origin, 95% confidence interval and 95% confidence limit for (i) neurochemical ratios measured with SVS vs. neurochemical ratios measured with MRSI quantified with the atlas-based (ATL) approach; (iI) neurochemical ratios measured with SVS vs. neurochemical ratios measured with MRSI and quantified with the ROI-based (ROI) approach; (iii) neurochemical ratios measured with SVS vs. neurochemical ratios measured with MRSI quantified with the four-voxel (4VX) approach. The regression equations and the R-squared (R2) of the regression for GABA+/tCr, tNAA/tCr and Glx/tCr are displayed in the plots accordingly.