Authors: Rachel C. Nethery, Alana M. Campbell and Young Truong
Abstract: Blind source separation (BSS) is applied to study brain connectivity. The conventional methods such as independent component analysis (ICA) address the problem by utilizing only the marginal information. An improvement has been suggested by exploiting the temporal or spatial correlation structure and has been demonstrated to be very important for extracting brain networks. A challenge for ICA has been the problem of assessing the significance of the independent components (IC) or latent sources. We propose to examine this problem by using a bootstrap method to resample the data. The procedure allows one to set confidence intervals for parameters or features related to the latent sources. An application based on EEG data will be illustrated to test if the source is truly a signal or noise.
Keywords:Independent component analysis, blind source separation, power spectrum, maximum likelihood, resting state, EEG, fMRI, connectivity, networks.