Evaluation of manual methods for integration of in vivo phosphorus NMR spectra

Eight human subjects measured peak integrals from a set of simulated 31P NMR spectra by triangulation, by extrapolation of tangents from the baseline of computed integrals, and by cursor demarcation of computed integrals using standard commercial software. The test spectra were qualitatively similar to phosphorus spectra routinely acquired from mammalian skeletal muscle. All three of the above methods were found to underestimate the areas of relatively broad peaks (e.g., ATP peaks were underestimated by 20-40%), even when the peak signal/rms-noise ratio was better than 8/1. This bias resulted in major errors in calculated metabolite ratios. Of the three methods, triangulation was the least accurate, and showed the greatest variation between subjects. Cursor demarcation of computed integrals was the most accurate manual method. The same test spectra were also integrated by two fully-automated fitting algorithms. Although not free of bias, these methods were generally superior to the manual methods.