MEGA-PRESS acquisitions to measure GABA+ (i.e. GABA and the co-edited macromolecule (MM) signal) are available on each of the three major vendors (GE, Siemens and Philips) at 3T; however, differences in sequence implementation (i.e., shape and bandwidths of slice selective pulses, duration, shape, bandwidth and timing of editing pulses, etc.) result in difference in GABA editing efficiency and MM co-editing. This limits comparisons of quantitative results across the literature, and large multi-site studies. Here, we characterize the sequence differences in terms of: (a) the editing efficiency of GABA; and (b) the co-editing of the macromolecules (MM), and assess the consistency of measurements across the vendors after correcting for these factors.

A phantom containing GABA and glycine with known T1 and T2 relaxation constants was used to determine the GABA editing efficiency - κ. Edited measurements with the following parameters: TR/TE = 2s/68 ms; 64 averages, 2048 datapoints; voxel size 8 cm³. For GE and Philips editing pulses were 14 ms and for Siemens editing pulses were 20 ms. Editing pulses were applied at 1.9 ppm (ON) and 7.5 ppm (OFF). The editing efficiency κ_i (where i is Philips, Siemens or GE) was calculated according to: $$ \kappa_i=\frac{I_{GABA,DIFF}e^{-TE/T_{2,Gly}}(1-e^{-TR/T_{1,Gly}})}{I_{Gly,OFF}e^{-TE/T_{2,GABA}}(1-e^{-TR/T_{1,GABA}})} $$ where I_GABA,DIFF is the integral of GABA in the averaged difference spectrum (i.e. the difference between then ON-spectrum and the OFF-spectrum, divided by two), I_Gly,OFF is the integral of glycine in the time-averaged OFF spectrum.

Relative co-editing of MM, μ was defined as the fraction of maximal, on-resonance editing that occurs at an offset of 0.2 ppm, (corresponding to the 1.7 ppm MM resonance) and determined acquiring an editing-pulse frequency series on each of the three scanners.

To correct for differences in GABA editing and MM co-editing, we assume that the GABA signal is proportional to κ; that the MM signal is proportional to κ and μ; and that 50% of the in vivo GABA+ signal acquired in the Philips implementation is MM [1]. Based on these assumptions the correction factor for inter-vendor differences is μ_Philips/κ_i(μ_Philips+μ_i), where i designates the vendor (GE, Philips or Siemens).

Two in vivo experiments were performed. Experiment 1 consisted of eight repeat scans of a single subject in each of the scanners. Experiment 2 consisted of 8 subjects scanned once in each of the scanners. These two experiments allow for intra- and inter-subject comparisons. For all acquisitions, a (3cm)³ voxel was placed in the sensorimotor cortex. Common acquisition parameters were: TR/TE = 2s/68 ms; 320 dynamics alternating ON-OFF every 2 averages. The editing pulses were 14 ms for GE and Philips and 15 ms for Siemens. Data were processed using Gannet 2.0 [2], applying 3 Hz line-broadening, spectral registration [3], and zero-filling out to 32k points. Baseline GABA levels were determined relative to water and Cr using the standard processing pipeline. κ,μ-corrected GABA levels using the vendor specific constants were then determined applying μ_Philips/κ_i(μ_Philips+μ_i) in place of the typical κ and MM correction factors [2]. This correction factor was also applied to GABA/Cr data.

For GE, Siemens and Philips, κ was determined to be: 0.45, 0.34 and 0.39, respectively and μ was measured to be: 0.83, 0.63 and 0.73, respectively. Due to poor data quality, two Siemens datasets from Experiment 1 and one Philips dataset from Experiment 2 were removed. Example spectra are shown in Figure 1. Baseline GABA and κ,μ-corrected GABA levels are compared in Figure 2. The across vendor coefficient of variation (CV) for the single subject decreases for the GABA/water data (baseline = 0.24, κ,μ-corrected = 0.21) and GABA/Cr data (baseline = 0.23, κ,μ-corrected 0.13). For the multiple subjects GABA/water CV increased (baseline = 0.067, κ,μ-corrected CV = 0.14) and the GABA/Cr CV decreased (baseline = 0.13, κ,μ-corrected = 0.049).

Differences in acquisition parameters to address competing demands in pulse sequence design has resulted in differences in the widely implemented MEGA-PRESS pulse sequences across the different vendors. We attempt to resolve these differences by defining the parameters κ and μ to characterize the differences in editing efficiency of GABA and the co-editing of MM. Overall, κ,μ-correction provides a moderate improvement in agreement among the different scanner platforms; however, differences still need to be resolved to enable comparisons between vendors.

Factors not captured in the κ,μ-correction such as B0 stability and shimming methods impact the underlying lineshape of the spectra, and modeling efficiency. Within and between subject reliability impacts the ability to quantify inter-scanner differences. For example, the different results between Experiment 1 and 2 GABA/water CVs suggests subject variability and compliance may confound results.