Dimension reduction is critical in identifying a small set of discriminative features that are predictive of behavior or cognition from high-dimensional neuroimaging data, such as EEG and fMRI. In the present study, we proposed a novel nonlinear supervised dimension reduction technique, named PCA-SIR (Principal Component Analysis and Sliced Inverse Regression), for analyzing high-dimensional EEG time-course data. Compared with conventional dimension reduction methods used for EEG, such as PCA and partial least-squares (PLS), the PCA-SIR method can make use of nonlinear relationship between class labels (i.e., behavioral or cognitive parameters) and predictors (i.e., EEG samples) to achieve the effective dimension reduction (e.d.r.) directions. We applied the new PCA-SIR method to predict the subjective pain perception (at a level ranging from 0 to 10) from single-trial laser-evoked EEG time courses. Experimental results on 96 subjects showed that reduced features by PCA-SIR can lead to significantly higher prediction accuracy than those by PCA and PLS. Therefore, PCA-SIR could be a promising supervised dimension reduction technique for multivariate pattern analysis of high-dimensional neuroimaging data.
展开▼
