Synopsis

Keywords: Data Analysis, Spectroscopy, GABA, MR Spectroscopy, MRSI, Reproducibility, Brain

Reproducibility of both conventional and GABA-edited multi-slice MRSI of the human brain recorded at 3T was investigated in 11 healthy adult volunteers. GABA-edited data were post-processed using a retrospective motion compensation (MoCo) scheme. Test-retest coefficients of variation(CV) were calculated. Overall reproducibility was good for conventional MRSI for selected brain regions, but edited-MRSI data showed significant variations. Edited-MRSI scans may benefit from both improved post-doc correction schemes to minimize subtraction artifacts, as well as better B₀ field homogeneity over the brain, and possibly also prospective MoCo schemes.

Introduction

Multi-slice MR spectroscopic imaging (MRSI) of the human brain at 3T has been used extensively over the past two decades for clinical and research investigations^1,2. However, there are only few previous studies of its reproducibility. More recently, techniques have been introduced for single- or multi-slice MRSI of GABA³, based on the use of spectral editing methodology, including retrospective motion compensation methods for reduction of subtraction artifacts in edited spectra⁴.
The purpose of this study was to evaluate the reproducibility of both conventional and edited multi-slice 2D MRSI measurements in the normal human brain at 3T, including the use of retrospective motion compensation.

Methods

11 healthy volunteers: 9 adults, 3M/6F, age 28 ± 6.7 years, min 22, max 43, and 2 children (1M/1F, age 7/11 years) were scanned twice between 7-14 days apart (mean ± st.dev. = 10±3 days).
An MR protocol including anatomical MRI, B₀ field maps for 2nd-order shim correction, conventional spin-echo and GABA-edited multi-slice MRSI, and a water reference MRSI was implemented. All experiments were performed on Philips 3T ‘Ingenia Elition’, 32-channel head coil with a total scan time of 45 minutes. Hypergeometric dual-band (HGDB) and outer-volume suppression (OVS) pulses water and lipid suppression⁵ were used for the conventional and edited-MRSI sequences. All MRSI scans were performed with three 15mm oblique-axial slices (2.5 mm gap, lateral ventricles to the vertex), 14x17 matrix (circular k-space sampling), nominal voxel size 12x12x15 mm (≈2.2 cm3). For GABA-editing, sequence parameters were TR/TE 1.8s/68ms, 4 excitations, scan time 22m 21s, 20 ms sinc-Gauss editing pulses (‘sg150’, 97 Hz bandwidth), edit ON/OFF frequencies 1.9/0.7 ppm. Parameters for conventional MRSI were TR/TE 1.7s/20ms, 1 excitation, scan time 5m 26s, and for water MRSI TR/TE 0.85s/20ms, 1 excitation, scan time 2m 39s.
The MRSI data were analyzed in Osprey⁶. For GABA-edited MRSI, retrospective MoCo was applied on raw k-space data prior to spatial Fast Fourier Transformation (FFT) as described previously⁵. After spatial transformation, coil combination and phase-correction were applied based on the water reference scan. Metabolite estimates were derived with linear-combination modelling using sequence-specific basis sets and quantified relative to total creatine (tCr). GABA-edited MRSI was analyzed with an optimized LCM approach for GABA-edited spectra⁷. Quantitative estimates from 14 selected brain regions over 3 slices (Figure 1) were analyzed including N-acetylaspartate (NAA), total-choline (tCho), glutamate plus glutamine (Glx), and myo-inositol (mI) estimates from the conventional MRSI and GABA+ from the GABA-edited MRSI Finally test-retest coefficients of variation (CV) were used to evaluate the reproducibility.

Results

Figure 1 shows the brain regions used for analysis. Figure 2 shows (A) conventional and (B) GABA-edited MRSI spectra for the 2 visits from a single subject, showing good spectral quality in terms of SNR, line-widths, and low residual water and lipid resonances. Subtraction artifacts (residual choline peak at 3.2 ppm) are seen in some of the GABA-edited MRSI spectra in either visit 1 or visit 2. Figure 3 shows the mean CVs for the 4 major metabolites (as described previously) measured in conventional MRSI, as well as mean CVs of GABA+ (GABA plus macromolecules), expressed as ratios to tCr for each brain region. As expected, CVs were lower for the compounds with larger, singlet signals (tNAA, tCho), with GABA+/tCr showing the largest CVs. Figure 4 shows mean tNAA/tCr, tCho/tCr, Glx/tCr, mI/tCr and GABA+/tCr ratios for all 14 brain regions at both study visits.

Discussion and Conclusions

Multi-slice conventional MRSI at 3T was found to be reproducible within 10 to 20% CV over the 1-to-2-week interval for the major metabolites in most brain regions. In contrast, GABA-edited MRSI data currently shows poor reproducibility in this study, with typical CVs in the 30 to 40% range for many brain regions, and as high as 80% in the worst case.
CVs were regionally dependent as expected, due to variations in B₀ field homogeneity, as well as lipid contamination from the scalp in more peripheral locations. Of note the genu of the corpus callosum (voxel 9 in figure 1) generally gave poor results because of susceptibility effects due to its proximity to sphenoid sinus. Peripheral brain regions also tended to have less good quality/reproducibility (compared to more central regions) due to worse B₀ field homogeneity and residual scalp lipid signals. Reasons for the worse reproducibility in the edited-MRSI spectra remain to be fully investigated, but presumably reflect the much smaller GABA+ signal (compared to the conventional metabolites), as well as incomplete removal of subtraction artifacts, which awaits further refinements.

Acknowledgements

Supported by NIH grants R01EB028259 and P41EB031771, and DOD W81XWH2010819.