3D GABA imaging with real-time motion correction, shim update and reacquisition of adiabatic spiral MRSI

Abstract

Gamma-aminobutyric acid (GABA) and glutamate (Glu) are the major neurotransmitters in the brain. They are crucial for the functioning of healthy brain and their alteration is a major mechanism in the pathophysiology of many neuro-psychiatric disorders. Magnetic resonance spectroscopy (MRS) is the only way to measure GABA and Glu non-invasively in vivo. GABA detection is particularly challenging and requires special MRS techniques. The most popular is MEscher-GArwood (MEGA) difference editing with single-voxel Point RESolved Spectroscopy (PRESS) localization. This technique has three major limitations: a) MEGA editing is a subtraction technique, hence is very sensitive to scanner instabilities and motion artifacts. b) PRESS is prone to localization errors at high fields (≥3T) that compromise accurate quantification. c) Single-voxel spectroscopy can (similar to a biopsy) only probe steady GABA and Glu levels in a single location at a time. To mitigate these problems, we implemented a 3D MEGA-editing MRS imaging sequence with the following three features: a) Real-time motion correction, dynamic shim updates, and selective reacquisition to eliminate subtraction artifacts due to scanner instabilities and subject motion. b) Localization by Adiabatic SElective Refocusing (LASER) to improve the localization accuracy and signal-to-noise ratio. c) K-space encoding via a weighted stack of spirals provides 3D metabolic mapping with flexible scan times. Simulations, phantom and in vivo experiments prove that our MEGA-LASER sequence enables 3D mapping of GABA+ and Glx (Glutamate+Gluatmine), by providing 1.66 times larger signal for the 3.02ppm multiplet of GABA+ compared to MEGA-PRESS, leading to clinically feasible scan times for 3D brain imaging. Hence, our sequence allows accurate and robust 3D-mapping of brain GABA+ and Glx levels to be performed at clinical 3T MR scanners for use in neuroscience and clinical applications.

Keywords: Frequency drift correction; GABA; Glutamate; LASER; MEGA editing; Magnetic resonance spectroscopy; Neurotransmitter; Prospective motion correction; Reacquisition; Real-time correction; Spiral imaging.

Fig. 1

Sequence schema of the MEGA-LASER sequence with 3D acquisition weighted stack of spirals encoding. Full adiabatic selection is provided by an AHP excitation pulse and three pairs of spatially-selective GOIA-W(16,4) refocusing pulses. Two frequency-selective Gauss refocusing pulses and adjacent spoiler gradients are used for MEGA editing. Phase encoding gradients in the z-direction are superimposed on top of the last MEGA spoiler. Constant-density spiral readout encode the x,y-plane. Preceding water suppression and volumetric navigator modules are not shown.

Fig. 2

Simulations for the GABA triplet at 3.01 ppm as observed in the subtraction spectra of the MEGA-LASER and MEGA-PRESS sequences. (a) Comparison of the effects of different positions and time intervals of the MEGA pulses in the MEGA-LASER sequence on the spectral appearance of the edited 3ppm GABA triplet. The MEGA pulses divide the total TE in three intervals (i.e., TE1/TE2/TE3 which include N1/N2/N3 GOIA pulses). The highest edited signal was obtained when the MEGA pulses were separated by half of the echo time (TE2=34 ms) in the second part of the echo (i.e., N1=3; N2=3; N3=0). (b) MEGA-LASER achieves ~2 times higher integrated GABA signal amplitude than MEGA-PRESS. The change in the GABA triplet pattern due to ±10% change in B1+ field was significantly smaller for (c) MEGA-LASER (3–7%) than for (d) MEGA-PRESS (12–28%). Slice profiles for PRESS and LASER pulses are shown in Supplementary Fig. 1.

Fig. 3

A cylindrical multi-compartment localization phantom (rotated by 45° around its axis) was used to illustrate the performance of the motion correction methods implemented in the MEGA-LASER sequence. The center of the sphere contains a rectangular compartment of 5×5×5 cm³ containing brain metabolites at close to physiological concentrations. The central compartment was surrounded by eight Falcon tubes filled with 10% ethanol or water mixed with Gd-DTPA to have enough image contrast to be traceable during motion. LCModel fits of five sample MEGA-edited spectra are displayed for the same voxel positioned inside the multi-compartment localization phantom (indicated by blue square). The top row (“motion”) shows LCmodel fits of spectra for data that were acquired during phantom movement. The bottom row (“static”) shows spectra obtained without phantom movement. The left column shows results without any correction (“noMoCo”), the center an example for using only B₀ shim and motion correction (“ShMoCo”), and the right column shows results, when selective reacquisition is used in addition to B₀ shim and motion correction (“ReShMoCo”). Spectra obtained during static condition without correction and with ReShMoCo correction were comparable. Spectra obtained during motion were significantly corrupted, if no correction was performed, and significantly improved when using ShMoCo. Further improvement in spectral quality was observed when using ReShMoCo. Substantial subtraction artifacts and contamination from signal outside the VOI are indicated by arrows. Plots of motion tracking are shown in Supplementary Fig. 2.

Fig. 4

Spectral grids illustrating the spatial heterogeneity of the subtracted GABA triplet (a–e) and Glu doublet (f–j) pattern that is observable in the central transversal slice obtained from a homogeneous spherical phantom (diameter 16 cm; 20 mM GABA and 20mM Glu solution). The B₁⁺ inhomogeneity was similar to conditions found in vivo (85±9° inside VOI). MEGA-editing via (a,f) LASER was compared to (b,g) PRESS using 3D-MRSI (FOV 20×20×16 cm³; VOI 10×10×6 cm³; matrix 20×20×16 interpolated to 32×32×16). The grid clearly shows improved spatial VOI selection for (a,f) MEGA-LASER compared to (b,g) MEGA-PRESS. Three representative spectra were selected from each grid (red rectangular boxes; c,h-top; d,i-center; e,j-bottom) for a direct comparison between MEGA-LASER (red lines) and MEGA-PRESS (black lines). In particular, they illustrate: (c) reduced GABA/Glu signal amplitude in PRESS due to the 4-compartment effect in the anterior direction; (d) altered signal intensity ratio between inner-to-outer GABA resonance lines due to too high B₁⁺ in the center of the phantom; and (e) lower GABA/Glu signal integral for MEGA-PRESS compared to MEGA-LASER even under ideal conditions. The overall GABA signal integral within the VOI (excluding border voxels) was 1.66 times larger for MEGA-LASER (in a.u. 43±5) than for MEGA-PRESS (26±5). For certain voxels the changes in the subtracted GABA integral were even larger (~2). Similar behavior can be observed for Glu. Note: Identical scaling was used for MEGA-LASER and MEGA-PRESS, but different scaling between GABA and Glu. A 6 Hz exponential filter was applied for display purposes.

Fig. 5

(a) Translation, (b) rotation, (c) first-order shim terms, and (d) frequency changes, as determined by the vNav and corrected during the 3D-MRSI scan (10×10×10 matrix) are plotted as a function of measurement time in the scanner coordinate system. The scan time of ~8 min was prolonged by ~2 min due to necessary reacquisition of corrupted data. During the scan, two different motion tasks were performed by the volunteer: (1) repeated head rotation in the transversal plane (right-left); and (2) repeated head rotation in the sagittal plane (up-down). Frequency changes shown in (d) illustrate the combined effect of head motion and scanner instability-related frequency drift.

Fig. 6

Five stacks of subtraction spectra of volunteer number 5 showing the frequency range from 2.8–4.2 ppm containing Glx (left signal at ~3.8ppm) and GABA+ (right signal at ~3ppm) for the VOI (see white rectangular box inside the brain) of a central transversal slice obtained via a 3D-MEGA-LASER acquisition with an 8 cm³ isotropic resolution in a healthy volunteer. In (a), the top row (“motion”) shows stacks of spectra for data that were acquired, while the subject was performing a head motion task. The bottom row (“static”) shows stacks of spectra obtained, while the subject was instructed not to move. The left column shows results without any correction (“noMoCo”), the center an example for using only B₀ shim and motion correction (“ShMoCo”), and the right column shows results, when selective reacquisition is used in addition to B₀ shim and motion correction (“ReShMoCo”). Without correction both static and motion scans were affected by subtraction artifacts. There was a lot of scanner instability causing subtraction artifacts even in the static uncorrected case. ShMoCo alone did not fully eliminate subtraction artifacts. In comparison, ReShMoCo significantly increased spectral quality for the static and even more in the motion case. (b) Sample LCModel fitting for spectra from the posterior cingulate (position indicated by red square) for all five different motion and correction methods. Note – the same scaling was used for all spectra and 3 Hz exponential filtering was applied to reduce noise level only for visualization purpose.

Fig. 7

Box plot illustrating the SNR of subtraction artifacts at ~3.2 ppm (Cho) of all spectra inside the VOI mask as a measure of contamination for different combinations of motion tasks and correction methods (i.e., motion task with no motion correction; motion task with shim- and motion correction; motion task with reacquisition, shim- and motion correction; static head position with no motion correction; and static head position with reacquisition, shim- and motion correction). An SNR threshold of SNR=3 was defined to filter contamination that can be considered noise. Only contamination with SNR > 3 was considered substantial and further processed.

Fig. 8

Percentage of contaminated voxels (i.e., defined as SNR of Cho subtraction artifact > 3) within the investigated mask displayed for all five volunteers (Vol#1–5) and for different combinations of motion tasks and correction methods (i.e., motion task with no motion correction; motion task with shim- and motion correction; motion task with reacquisition, shim- and motion correction; static head position with no motion correction; and static head position with reacquisition, shim- and motion correction). Scans with ReShMoCo had significantly lower percentage of contaminated voxels than ShMoCo or no correction.

Fig. 9

Morphological T1-weighted reference images (left), 3D GABA+ maps obtained by MEGA-LASER and ReShMoCo in a healthy volunteer with 1 cm³ isotropic resolutions in ~24 min (center), and color-coded overlay (right) displayed in transversal, sagittal, and coronal plane. For display purposes GABA+ maps were interpolated to the T₁-weighted MRI. The contours of the ventricles, as well as increased GABA+ levels in regions that predominantly contain grey matter are well visible on the 3D GABA+ maps as visible from direct comparison with morphological reference images. Detailed imaging parameters are listed in table 1.