Independent modulations of high-gamma band spectral activity in human scalp EEG distinct from scalp muscle activity

Julie Onton & Scott Makeig

Institute for Neural Computation

University of California San Diego

INTRODUCTION: From the earliest observations of scalp and cortical EEG activity, marked changes in the power spectrum linked to changes in attention and awareness have been noted, slow frequencies predominating in sleep and high frequencies associated with focused attention. However, during even continuous task performance, frequent spectral shifts occur in many parts of cortex. We have recently developed a method for separating different modes of spectral modulation in scalp EEG or other electrophysiological data. This method uses infomax independent component analysis (ICA) to extract independent modulator (IM) processes from normalized spectrograms of concurrent independent component (IC) processes. First application to scalp data from 34 subjects participating in an emotion imagination task produced clusters of IMs whose frequency templates were either periodic with harmonics, roughly periodic without harmonics, or broadband. IMs in the latter category had a high-pass character, not returning to baseline by 50 Hz, the top of our analysis window. We therefore reproduced this analysis using a wider frequency band (1-125 Hz). To check the possibility that the wideband IMs merely represented spatial leakage from scalp muscle activity, we also included ICs accounting for single scalp muscle activities in the analysis.

METHODS: Using a moving-window Fourier transform we computed log spectral transforms of 50%-overlapping 1-s epochs of the activities of ICs with near-dipolar maps from continuous 100-channel data collected during an emotion imagination task. The resulting 1-125 Hz IC log spectrograms were normalized and then jointly decomposed by infomax ICA, returning for each subject 15 independent modulator (IM) factors, each with a frequency template for each IC included in the analysis.

RESULTS: Figure 1 shows a portion of a typical wide-band IM decomposition. Scalp maps (top) show the included ICs, left boxes the window-weight histograms. Each row of traces show spectral effect templates of one IM process on each IC. Several (upper) IMs account for changes in electromyographic (EMG) activity levels in (left) ICs accounting for EMG in single scalp muscles—note their expected (20-100+ Hz) EMG profiles. These IMs, however, are not linked to spectral changes in (right) cortical ICs. The cortical broadband (lower) IMs, likewise not linked to EMG shifts, exhibit two patterns, either still increasing or flattening at 125 Hz or peaking below 100 Hz (bilateral inferior frontal IC9). One IM that co-modulates several ICs is shown (bottom row).

DISCUSSION: Recently, high-gamma band (roughly 80-200 Hz) activity has been discovered in human intracranial field recordings. These results indicate that spectral shifts including the high-gamma band are present in non-invasively obtained scalp EEG data, and that these cannot be explained by contamination from scalp EMG. Independent modulator analysis (IMA) reveals patterns of broadband cortical activity, among others, whose function and functional control systems.