The EEG recordings were re-referenced to the average of TP9 and TP10 channels, positioned close to the subject's mastoids (digitally linked mastoids reference). A modified version of the adaptive artifact subtraction (AAS) algorithm was used for off-line correction of imaging artifact (Allen et al., 2000; Gonçalves et al., 2007): after the detection of each fMRI slice onset from EEG data, slice artifact waveforms were segmented, averaged, and iteratively subtracted from the EEG signals (Gonçalves et al., 2007). Subsequently, the EEG data were downsampled to 1 kHz and filtered between 0.5 and 40 Hz by means of a Chebychev II-type filter with 40 dB attenuation and zerophase distortion. Simulated signals for ballistocardiographic (BCG), slice MRI and volume MRI artifacts were generated, using information from acquired data. An average BCG waveform was constructed by differentiating the EKG signal, detecting QRS peak timing, and averaging with respect to QRS peaks. The BCG template was then obtained by replicating the same averaged waveform across heart beats, shifted by a fixed delay of 150 ms in order to take into account the typical difference in
