In EEG-fMRI studies, ICA was previously proposed by Calhoun, for direct ERP-fMRI fusion [Calhoun et al., 2006], and by Eichele, for single-trial EEG-fMRI matching [Eichele et al., 2008]. In particular, Calhoun suggested to fuse the electromagnetic temporal information in the ERP with data the hemodynamic spatial information in the fMRI data by using the joint constraints of ERP and fMRI independence in the ICA model. This approach, although proven to be valuable in the study of time-locked activity, might not be suitable to provide a description of the activation timing across distinct brain networks. Alternatively, Eichele proposed to perform ICA separately on EEG and fMRI data, and to match the ICs that are consistent across modalities on the basis of their trial-to-trial variations. This approach may reveal interesting correspondences between electrophysiological and hemodynamic measures [Heeger and Ress, 2002]; on the other hand, it might be sensitive to an incomplete matching between the EEG and fMRI time-courses associated with the same generators.
