Guro Fannelb Giskedegrd1, Tom Bloemberg2, Lutgarde Buydens2, Geert Postma2, Ingrid Susanne Gribbestad1, Tone Frost Bathen1
1Dept. of Circulation and Medical Imaging, Norwegian University of Science and Technology (NTNU), Trondheim, Norway; 2Dept. of Analytical Chemistry, Radboud University Nijmegen, Nijmegen, Netherlands
Correction of misaligned peaks is an important part of multivariate preprocessing of MR spectra. In this study, three different peak alignment algorithms were tested on HR MAS MRS data from breast cancer tissue. The datasets were used to predict the prognostic factor ER status, which is shown to be related to metabolic profile. Correlation optimized warping (COW) and peak alignment by genetic algorithm (PAGA) resulted in greatly improved PLS-DA classification of ER status compared to unaligned data. Parametric time warping (PTW) did not improve the classification error, indicating that PTW may not be as suitable for metabolomic MR data.