Blind source separation of multiple signal sources of fMRI data sets using independent component analysis

Purpose. As we seek to establish the use of fMRI as the method of choice for studying systems-level neuroscience, it is essential that we be able to distinguish the various “signal ” sources from the “noise ” sources in which they are immersed. Almost all published fMRI studies thus far have been used to localize signal changes related to brain activity but not to determine the source of the individual signals. A better understanding of these signal sources could not only lead to improved reliability and SNR of task-induced signal changes, but could also improve our understanding of the biophysical basis of signal origin. In this study, we apply a blind-source separation technique to determine multiple signal sources in FMRI time-series data sets and compare this analysis method to PCA in the same data sets. Methods. SeeRef. 1 for details ofICAmethodology