MRS data analysis

MRS data analysis was performed by using the Java-based Magnetic Resonance User Interface (jMRUI) v. 5.0 (http://www.mrui.uab.es/mrui/). The jMRUI software package enables the time-domain analysis of in vivo MRS data in two stages: preprocessing and quantization. Preprocessing techniques included eddy current compensation, Hankel–Lanczos singular value decomposition filtering (HLSVD), offset correction, zero filling, apodization, phase correction, and baseline correction (Mandal, 2012). Before fitting, spectra were preprocessed automatically by eddy current correction and phase coherent frequency shift correction, resulting to an improvement of the signal-to-noise ratio (SNR). Pre-processing requires user interaction using the HLSVD filter largely to suppress residual water molecules, and use of the Cadzow function to filter the signal. The quantities of interest were calculated with the Advanced Method for Accurate, Robust, and Efficient Spectral fitting (AMARES) (Vanhamme et al., 1997). AMARES fitting ensures the incorporation of more prior knowledge on the spectral parameters to increase efficiency, overall accuracy, and, convergence rates and can also be extended to fit echo signals. AMARES is preprogrammed to switch between the Lorentizian, Gaussian, and Voigt models, and can be used for fitting spin echoes in addition to free induction decay (Naressi et al., 2001; Mandal, 2012). Spectral fitting used Lorentzian line shapes. In order to improve the quantification process, this method depends on a prior knowledge entered by the user of the sought resonance peaks. A range of peak line width for the variation was also entered. These were allowed to vary between 2 and 14 Hz. The same prior knowledge to estimate peaks set at the following positions: 2.02 parts per million (ppm) and 3.9 line width [LW(Hz)] for N-acetylaspartate, 2.35 ppm and 4.9 LW for glutamate, 3.01 ppm and 4.9 LW for creatine, 3.2 ppm and 4.9 LW for choline (Cuellar-Baena et al., 2011; Scott et al., 2016). MRS data processed using AMARES in the jMRUI software package provided information about the estimated components, including frequencies, damping, amplitudes, and phases. Cramer–Rao lower bounds (CRLBs) were used as a measure of the accuracy of a calculation of the amplitude of a certain component. Only metabolite concentrations with CRLBs below 20% were accepted and used in subsequent analyses. The concentrations of metabolites were calculated according to as described in detail previously (Helms, 2008), as follows

where C stands for concentration, S for signal intensity, V for size of the VOI, R for the receiver gain, and ext for external reference.

Numerous effects should be taken into account, when performing MRS on a phantom, and getting absolute concentration estimates requires a calibration. First, the measurement on phantom is generally performed at a lower temperature (Temp) than the human body temperature, so yielding a polarization increased by 310 K/Temp. The relaxation times, T₁ and T₂, which are normally somewhat longer in vitro than in vivo, will also be impacted by the temperature. Scaling the reference signal to match the conditions in vivo should be considered. Additionally, a mismatch of coil impedance along with the associated scaling and reflection losses, due to the load of the phantom is small, as described in detail previously (Helms, 2008).

where S stands for signal intensity, Temp for temperature of the phantom, and ext for external reference. The results for the metabolites of interest were then corrected for partial volume and relaxation effects as an average of the metabolites of interest reported in the literature. Briefly, for N-acetylaspartate, T₁ = 1.47 s and T₂ = 247 ms; for choline, T₁ = 1.30 s and T₂ = 207 ms; for creatine, T₁ = 1.46 s and T₂ = 152 ms; for glutamate, T₁ = 1.27 s and T₂ = 199 ms (Mlynárik et al., 2001; Mullins et al., 2008).