Baseline correction of functional MR time courses with PCA

Abstract

Statistical methods have been widely utilized for analyzing data observed in experiments using functional magnetic resonance imaging (fMRI) techniques. In confirmatory experiments, for example, the desirable voxel time courses are known and correlation analyses, tor nonparametric tests and time frequency analyses can be used to identify activation areas that are task relevant and scientifically meaningful. The fMRI time courses may be contaminated by a drift in baseline signal within and across different experimental runs due to subjects' movement and instrumental instability. In the literature, the baseline variation is corrected using the global normalization or linear detrending methods. Drifts in time courses normally appear as major principal components in voxel intensity. In this study, we applied the principal component analysis (PCA) technique to removing baseline artifacts and other noises from fMRI time courses. The correlation between corrected voxel intensity and a reference function was analyzed to identify task-related components, and was compared with those using the standard detrend method. The results suggest that the PCA method gives higher correlation coefficients on average than the standard detrending method.