Our EEG–fMRI method based on ICA for fMRI analysis allowed linking information from EEG and fMRI data according to a data-driven approach. Therefore, it can be assumed to be unbiased by the use of an incomplete model for the fusion of EEG and fMRI data. For example, although we extracted only the most prominent feature from the EEG data (the P300 component), we could relate the P300 response variations with the network ongoing activity. Conversely, the use of the P300 reference time-course in a hypothesis-driven EEG-based GLM analysis would not generally provide complete information, because it would strictly require, instead, a set of EEG predictors capable to comprehensively account for the fMRI time-courses. The present study showed that the EEG-informed fMRI analysis can be effectively conducted by ICA of fMRI data, just as previous studies by Debener showed that it can benefit from ICA of EEG data (Debener et al., 2005, 2006). The next logical step would be to use ICA on both EEG and fMRI. From this standpoint, Eichele suggested to use ICA in parallel on simultaneously acquired EEG
