Utility of Independent Component Analysis for Interpretation of Intracranial EEG

Diane Whitmer,Gregory Worrell,Il Keun Lee,Matt Stead,Scott Makeig

doi:10.3389/fnhum.2010.00184

Diane Whitmer, Gregory Worrell + Show 3 more

Open Access

https://doi.org/10.3389/fnhum.2010.00184

Copy DOI

Abstract

Electrode arrays are sometimes implanted in the brains of patients with intractable epilepsy to better localize seizure foci before epilepsy surgery. Analysis of intracranial EEG (iEEG) recordings is typically performed in the electrode channel domain without explicit separation of the sources that generate the signals. However, intracranial EEG signals, like scalp EEG signals, could be linear mixtures of local activity and volume-conducted activity arising in multiple source areas. Independent component analysis (ICA) has recently been applied to scalp EEG data, and shown to separate the signal mixtures into independently generated brain and non-brain source signals. Here, we applied ICA to unmix source signals from intracranial EEG recordings from four epilepsy patients during a visually cued finger movement task in the presence of background pathological brain activity. This ICA decomposition demonstrated that the iEEG recordings were not maximally independent, but rather are linear mixtures of activity from multiple sources. Many of the independent component (IC) projections to the iEEG recording grid were consistent with sources from single brain regions, including components exhibiting classic movement-related dynamics. Notably, the largest IC projection to each channel accounted for no more than 20–80% of the channel signal variance, implying that in general intracranial recordings cannot be accurately interpreted as recordings of independent brain sources. These results suggest that ICA can be used to identify and monitor major field sources of local and distributed functional networks generating iEEG data. ICA decomposition methods are useful for improving the fidelity of source signals of interest, likely including distinguishing the sources of pathological brain activity.

Highlights

The goal of cognitive neuroscience is to correlate behavior and experience with brain function
Accounted for 89% of the activity on Grid24 (Figure 9), and demonstrates the same alpha, beta, and gamma band movement-related spectral changes seen on Grid24 (Figure 9). To our knowledge, this is the first study in which Independent component analysis (ICA) has been applied to human intracranial data for interpretation of functional brain signals, Hu et al (2007) applied ICA to concurrently recorded intracranial EEG (iEEG) and scalp EEG to remove the contributions of sources near the scalp reference channel from the iEEG channel data
We applied ICA to intracranial data recorded during a cognitive task because it has been instructive for other researchers www.frontiersin.org in decomposing the linear mixtures recorded by scalp EEG data into maximally independent model sources whose dynamics are modulated on a task basis

Summary

Introduction

The goal of cognitive neuroscience is to correlate behavior and experience with brain function. While animal studies provide the opportunity to record neural activity directly, either at the level of single units or of local field potentials, human neuroscience is typically limited to non-invasive, whole-brain imaging techniques typically considered to have limited spatial and/or temporal resolution. Scalp electroencephalographic (EEG) recordings have submillisecond temporal resolution, but their spatial resolution is limited by the typical multi-centimeter scale spacing of electrodes on the scalp surface, by the broad point-spread functions of far-field potentials generated in cortical areas, and by the difficulty of estimating source distributions on the highly folded brain surface from sparse measurements on the smooth and electrically distant scalp surface. Estimation of the locations of neurophysiological current sources that generate the electric fields recorded by EEG sensors is inherently a mathematically underdetermined problem whose solution requires the use of additional physiological constraints to limit the infinite solution space. Many IC scalp maps visualizing their topographic projections to the EEG electrode montage exhibit biophysically simple patterns consistent with fields generated by dipolar current sources, in the absence of any explicit field pattern constraints in the ICA model (Makeig et al, 1997, 2004a)

Methods

Results

Conclusion