1H-MRS processing parameters affect metabolite quantification: The urgent need for uniform and transparent standardization

Abstract

Proton magnetic resonance spectroscopy (1H-MRS) can be used to quantify in vivo metabolite levels, such as lactate, γ-aminobutyric acid (GABA) and glutamate (Glu). However, there are considerable analysis choices which can alter the accuracy or precision of 1H-MRS metabolite quantification. It is currently unknown to what extent variations in the analysis pipeline used to quantify 1H-MRS data affect outcomes. The purpose of this study was to evaluate whether the quantification of identical 1H-MRS scans across independent and experienced research groups would yield comparable results. We investigated the influence of model parameters and spectral quantification software on fitted metabolite concentration values. Sixty spectra in 30 individuals (repeated measures) were acquired using a 7-T MRI scanner. Data were processed by four independent research groups with the freedom to choose their own individualized and optimal parameter settings using LCModel software. Data were processed a second time in one group using an independent software package (NMRWizard) for an additional comparison with a different post-processing platform. Correlations across research groups of the ratio between the highest and, arguably, the most relevant resonances for neurotransmission [N-acetyl aspartate (NAA), N-acetyl aspartyl glutamate (NAAG) and Glu] over the total creatine [creatine (Cr) + phosphocreatine (PCr)] concentration, using Pearson's product-moment correlation coefficient (r), were calculated. Mean inter-group correlations using LCModel software were 0.87, 0.88 and 0.77 for NAA/Cr + PCr, NAA + NAAG/Cr + PCr and Glu/Cr + PCr, respectively. The mean correlations when comparing NMRWizard results with LCModel fitting results at University Medical Center Utrecht (UMCU) were 0.87, 0.89 and 0.71 for NAA/Cr + PCr, NAA + NAAG/Cr + PCr and Glu/Cr + PCr, respectively. Metabolite quantification using identical 1H-MRS data was influenced by processing parameters, basis sets and software choice. Locally preferred processing choices affected metabolite quantification, even when using identical software. Our results reinforce the notion that standard practices should be established to regularize outcomes of 1H-MRS studies, and that basis sets used for processing should be made available to the scientific community.