Accurate quantification of the proton NMR spectra of human brain metabolites
Magnetic Resonance Spectroscopy (MRS) can give chemically specific information about biological tissue and has become an active area of research in human metabolism in health and disease. In particular, the (1)H MKS modality is suited to studies of the brain, and the commonly studied metabolites that have important biochemical roles are choline, creative and N-acetyl aspartate (NAA). The success of MRS as a medical diagnostic tool depends heavily on accurate quantification of the data by computer analysis, in order to determine frequencies for the identification of metabolites and peak areas to obtain metabolite concentrations. MRS performed on standard clinical scanners has problems regarding the homogeneity of the magnetic field across the region being measured and this results in a broadening of the inherently Lorentzian peaks of metabolite resonances. T'he broadening effect can be approximated by a Gaussian function and so observed peaks are of a Voigt form. However, traditional computer modeling of MRS data usually assumes a purely Lorentzian model. Although the Voigt function is theoretically the most appropriate model for quantification, its complex form presents an unnecessary time burden when analyzing multiple spectra. An accurate approximation to the Voigt function, consisting of a linear summation of Lorentzian and Gaussian functions, is given in this work. The maximum error in determining peak areas using the approximation is shown to be 0.72% over the entire range of simulated Voigt line shapes. Furthermore, it is shown that the approximation can be derived analytically from the actual Voigt function in the regions where the Lorentzian or Gaussian components are dominant. The (1)H spectra of dilute solutions containing choline, creative and sodium acetate were collected from an Elscint Prestige 1.9T and a GE Sigma EchoSpeed 1.ST clinical scanner and quantified by fitting with model line shapes using a Levenberg-Marquardt non-linear least squares routine. Compared with an approximated Voigt model, fitting with purely Gaussian and Lorentzian models resulted in a increase of (% mean ± SE) that was as much as 19.06 ± 0.39 and 405.48 ± 5.66 respectively. Similar trends were observed from fitting two in vivo spectra of a healthy volunteer, where using purely Gaussian and Lorentzian models resulted in a % increase of that was around 3% and 20% respectively. The relaxation times T(1) and T(2) of choline, creative and sodium acetate were estimated from measurements performed on the two clinical scanners and compared with estunates from similar measurements performed on a high resolution spectrometer. After correcting spectra for relaxation effects, the choline/creative and the choline/sodium acetate peak area ratios (mean ± SE) were 3.01 ± 0.02 and 3.03 ± 0.03 respectively on the GE scanner compared to an ideal value of 3. Use of an inappropriate line shape model leads to systematic errors not only in peak areas, but also area ratios of overlapping spectral peaks. Compared to fitting with the approximated Voigt model, noticeable discrepancies in the choline/creative area ratio was observed when fitting with the Gaussian model (3.7%) and the Lorentzian model (15.7%). Similarly for the in vivo spectra, a noticeable discrepancy in the NAA/creatine area ratio was observed when fitting with the Gaussian model (4.8%) and the Lorentzian model (12.6%). Therefore the approximated Voigt model returned more accurate fits and should be considered as the standard line shape in the reporting of spectroscopic studies of human brain metabolites. A small scale reproducibility study was performed on the two clinical scanners to determined the precision to which metabolite peak areas could be measured, and greater precision was observed on the GE scanner. The coefficient of variation (CV) of the `within-run' reproducibility (no repositioning of solutions between measurements) was within 3.7% for spectra of dilute solutions and 7.8% for in vivo spectra. The `within session' reproducibility (repositioning of sample between measurements) was within 0.9% for spectra of dilute solutions. The `between-days' reproducibility for spectra of dilute solutions, quantified by a t-test, showed no significant differences between both the mean peak areas for any of the metabolites or the choline/creative and the choline/sodium acetate peak area ratios (p > 0.05 in all cases).