How to perform automated curve fitting toin vivo ³¹P magnetic resonance spectroscopic data

Abstract

A procedure is presented for the computerized automated curve fitting ofin vivo ³¹P NMR data. This procedure was implemented in the form of three C shell scripts (Appendix) which automatically execute commands from the commercial software program, NMR1™. The accuracy and limitations of curve fitting was tested using simulated data designed to representin vivo ³¹P NMR spectra obtained from brain. For isolated peaks, the predicted areas for 140 test spectra were in good agreement with the noise free or ‘true’ values, with variations on the order of that expected for the calculated S/N of the simulated peaks. However, when the S/N was less than 2:1, predicted areas were systemically overestimated; this error was traced to a bias for linewidth overestimates. For peaks that overlap, a second systematic error was noted in predicted areas for adjacent peaks, where one peak area was overestimated and the other was underestimated. Furthermore, these systematic errors show partial inverse co-linearity with each other, increasing in proportion to the extent of peak overlap. The curve fitting procedure and tests described here provide guidelines and cautions to investigators who endeavour to use computerized procedures for the analysis ofin vivo NMR spectroscopic data using NMR1 or other software programs.