Figure 2.

Discriminatory EEG components and parametric EEG regressors. A, The area under the receiver operating curve (AUC) plotted as a function of time window relative to stimulus onset, for a linear classifier trained to discriminate between faces and nonfaces given the multichannel EEG. Shown are time courses of the AUC for the low-coherence and high-coherence stimulus levels, respectively, averaged across subjects. The shaded areas around the time courses indicate the SEM, while the dotted line represents the significance threshold at p < 0.05 (false discovery rate corrected) for the mean AUC, determined by a nonparametric permutation technique. The stars indicate significant time bins in the early and late windows. Also shown are the forward models for the EEG components. B, Illustration of EEG-informed fMRI analysis. In the general linear model analysis applied to the fMRI, four EEG regressors were included as BOLD predictors. They were constructed from the early and late components at the high-coherence and low-coherence levels, respectively. The onset time of the regressors matched the timing of each stimulus presentation. The amplitude of the EEG regressors was modulated by the classifier output on both face and nonface trials (trial-to-trial variability).