Application of self-organising maps in automated chemical shift correction of in vivo 1H MR spectra

Abstract

Frequency shift differences in1H MRSI spectra due to magnetic field inhomogeneities pose a problem, if automated lineshape fitting routines (LF) or artificial neural network (ANN) methods are used for spectral quantification. Use of self-organizing map (SOM) analysis for automated shift correction of long echo time (TE=270 ms) in vivo1H NMR spectra of human brain is demonstrated. The map is obtained by training a SOM with proton spectra and the chemical shifts of the reference vectors were calibrated. The maps were then used for classification of spectroscopic imaging data and the calibration information for corrections of chemical shifts.