Dynamic EEG-informed fMRI modeling of the pain matrix using 20-ms root mean square segments

Abstract

Previous studies on the spatio-temporal dynamics of cortical pain processing using electroencephalography (EEG), magnetoencephalography (MEG), or intracranial recordings point towards a high degree of parallelism, e.g. parallel instead of sequential activation of primary and secondary somatosensory areas or simultaneous activation of somatosensory areas and the mid-cingulate cortex. However, because of the inverse problem, EEG and MEG provide only limited spatial resolution and certainty about the generators of cortical pain-induced electromagnetic activity, especially when multiple sources are simultaneously active. On the other hand, intracranial recordings are invasive and do not provide whole-brain coverage. In this study, we thought to investigate the spatio-temporal dynamics of cortical pain processing in 10 healthy subjects using simultaneous EEG/functional magnetic resonance imaging (fMRI). Voltages of 20 ms segments of the EEG root mean square (a global, largely reference-free measure of event-related EEG activity) in a time window 0-400 ms poststimulus were used to model trial-to-trial fluctuations in the fMRI blood oxygen level dependent (BOLD) signal. EEG-derived regressors explained additional variance in the BOLD signal from 140 ms poststimulus onward. According to this analysis, the contralateral parietal operculum was the first cortical area to become activated upon painful laser stimulation. The activation pattern in BOLD analyses informed by subsequent EEG-time windows suggests largely parallel signal processing in the bilateral operculo-insular and mid-cingulate cortices. In that regard, our data are in line with previous reports. However, the approach presented here is noninvasive and bypasses the inverse problem using only temporal information from the EEG.

Figure 1

Averaged EEG response to laser stimulation. (a) RMS (root mean square) grand average. The arrows reflect the time windows typically reported for the N1, N2, and P2 peaks of the scalp‐recorded laser‐evoked potential [e.g. Kakigi et al., 2005]. Black line = grand average. Dashed red line = standard deviation. (b) Grand average of the laser‐evoked potential obtained from electrode position Cz. Black line = grand average. Dashed red line = standard deviation. (c) Field maps of three selected time windows.

Figure 2

BOLD response to laser stimulation, standard model. Second‐level mixed‐effects FLAME. N = 10 subjects. Cluster‐corrected threshold Z = 2.3, P = 0.01. Upper row: surface projection. Lower row: axial slices. R = right. L = left.

Figure 3

BOLD response to laser stimulation, EEG RMS‐informed model. Second‐level mixed‐effects FLAME. N = 10 subjects. Cluster‐corrected threshold Z = 2.3, P = 0.01. The additional activation that was specific for the regressor containing single‐trial EEG information is given for three EEG time windows poststimulus: from top to bottom 140−160 ms, 160−180 ms, and 300−320 ms poststimulus. R = right. L = left.