Quantification of human brain metabolites from in Vivo 1 H NMR magnitude spectra using automated artificial neural network analysis

Long echo time (TE =270 ms) in vivo proton NMR spectra resembling human brain metabolite patterns were simulated for lineshape fitting (LF) and quantitative artificial neural network (ANN) analyses. A set of experimental in vivo 1 H NMR spectra were first analyzed by the LF method to match the signal-to-noise ratios and linewidths of simulated spectra to those in the experi-mental data. The performance of constructed ANNs was compared for the peak area determinations of choline-containing compounds (Cho), total creatine (Cr), and N-acetyl aspartate (NAA) signals using both manually phase-corrected and magnitude spectra as in-puts. The peak area data from ANN and LF analyses for simulated spectra yielded high correlation coefficients demonstrating that the peak areas quantified with ANNgave similar results as LF analysis. Thus, a fully automated ANN method based on magnitude spectra has demonstrated potential for quantification of in vivo metabolites from long echo time spectroscopic imaging