Non-invasive detection of high gamma band activity during motor imagery

Melissa M. Smith, Kurt E. Weaver, Thomas J. Grabowski, Rajesh P. N. Rao, Felix DarvasFrontiers in Human Neuroscience 2014 · DOI

Abstract

High gamma oscillations (70-150 Hz; HG) are rapidly evolving, spatially localized neurophysiological signals that are believed to be the best representative signature of engaged neural populations. The HG band has been best characterized from invasive electrophysiological approaches such as electrocorticography because of the increased signal-to-noise ratio that results when by-passing the scalp and skull. Despite the recent observation that HG activity can be detected non-invasively by electroencephalography (EEG), it is unclear to what extent EEG can accurately resolve the spatial distribution of HG signals during active task engagement. We have overcome some of the limitations inherent to acquiring HG signals across the scalp by utilizing individual head anatomy in combination with an inverse modeling method. We applied a linearly constrained minimum variance (LCMV) beamformer method on EEG data during a motor imagery paradigm to extract a time-frequency spectrogram at every voxel location on the cortex. To confirm spatially distributed patterns of HG responses, we contrasted overlapping maps of the EEG HG signal with blood oxygen level dependence (BOLD) functional magnetic resonance imaging (fMRI) data acquired from the same set of neurologically normal subjects during a separate session. We show that scalp-based HG band activity detected by EEG during motor imagery spatially co-localizes with BOLD fMRI data. Taken together, these results suggest that EEG can accurately resolve spatially specific estimates of local cortical high frequency signals, potentially opening an avenue for non-invasive measurement of HG potentials from diverse sets of neurologically impaired populations for diagnostic and therapeutic purposes.

Introduction

In the field of neuroimaging, high gamma (HG) oscillations (70–150 Hz) are considered the "gold standard" for tracking active neural populations due to their spatial precision and rapid response to tasks. However, recording these signals non-invasively via electroencephalography (EEG) is difficult because the skull and scalp act as low-pass filters, significantly attenuating high-frequency signals. Furthermore, the HG frequency range overlaps with electromyographic (EMG) noise from the jaw and neck.

Traditionally, HG activity has been recorded using invasive electrocorticography (ECoG). This study aims to demonstrate that non-invasive EEG, when combined with individual anatomical modeling and advanced beamforming algorithms, can accurately resolve these localized signals during motor imagery tasks.

Methods

The researchers employed a multi-modal approach involving 10 neurologically normal adult subjects (mean age 24.9). The experimental task required subjects to imagine a finger-tapping sequence with either their left or right hand.

Data collection involved two primary modalities:

  1. EEG Source Imaging: 54-channel EEG was recorded at 1200 Hz. Individual head models (Boundary Element Models) were created from each subject’s structural MRI to account for unique anatomy.
  2. fMRI Validation: Subjects performed the same task in a 3T fMRI scanner to provide a spatial benchmark (BOLD signals) for comparison with the EEG HG activity.

To map the signals, the team applied a Linearly Constrained Minimum Variance (LCMV) beamformer. This acts as a spatial filter, suppressing noise from outside a specific voxel to extract time-frequency spectrograms at every cortical location.

Figure 1: Spatial distribution of EEG electrodes mapped onto a representative cortical surface used for source localization.

Key Results

The study successfully demonstrated that scalp-based HG activity spatially co-localizes with fMRI BOLD data, providing evidence of its genuine cortical origin.

  • Spatial Accuracy: The average distance between EEG HG clusters and fMRI BOLD clusters was 14 mm (SD 8 mm), representing a high degree of spatial agreement.
  • Temporal Consistency: Significant HG increases were consistently found between 0.3 and 1.0 seconds after the instruction to imagine movement.
  • Frequency Characteristics: HG activity typically appeared in narrow bands (often 70–90 Hz) rather than a broad spectrum.
  • Subject Reliability: Significant HG clusters were identified in 9 out of 10 subjects for right-hand imagery and 8 out of 10 for left-hand imagery.

Discussion and Future Work

These findings suggest that EEG can resolve spatially specific cortical signals, potentially enabling high-performance, non-invasive brain-computer interfaces (BCIs). Unlike traditional Mu or Beta rhythms, HG signals are more focal and evolve rapidly, which could lead to BCIs with higher information transfer rates.

However, limitations remain. The signal-to-noise ratio (SNR) for HG on the scalp is low, requiring the averaging of many trials (approx. 100 per hand). Additionally, individual variability in peak frequency and location means systems must be calibrated to the user. Future work will focus on testing these methods in clinical populations, such as stroke or ALS patients, to evaluate their potential for rehabilitation and communication.

Summary updated

Abstract

High gamma oscillations (70-150 Hz; HG) are rapidly evolving, spatially localized neurophysiological signals that are believed to be the best representative signature of engaged neural populations. The HG band has been best characterized from invasive electrophysiological approaches such as electrocorticography because of the increased signal-to-noise ratio that results when by-passing the scalp and skull. Despite the recent observation that HG activity can be detected non-invasively by electroencephalography (EEG), it is unclear to what extent EEG can accurately resolve the spatial distribution of HG signals during active task engagement. We have overcome some of the limitations inherent to acquiring HG signals across the scalp by utilizing individual head anatomy in combination with an inverse modeling method. We applied a linearly constrained minimum variance (LCMV) beamformer method on EEG data during a motor imagery paradigm to extract a time-frequency spectrogram at every voxel location on the cortex. To confirm spatially distributed patterns of HG responses, we contrasted overlapping maps of the EEG HG signal with blood oxygen level dependence (BOLD) functional magnetic resonance imaging (fMRI) data acquired from the same set of neurologically normal subjects during a separate session. We show that scalp-based HG band activity detected by EEG during motor imagery spatially co-localizes with BOLD fMRI data. Taken together, these results suggest that EEG can accurately resolve spatially specific estimates of local cortical high frequency signals, potentially opening an avenue for non-invasive measurement of HG potentials from diverse sets of neurologically impaired populations for diagnostic and therapeutic purposes.
BibTeX
@inproceedings{smith2014non,
  author = {Melissa M. Smith and Kurt E. Weaver and Thomas J. Grabowski and Rajesh P. N. Rao and Felix Darvas},
  title = {{Non-invasive detection of high gamma band activity during motor imagery}},
  booktitle = {Frontiers in Human Neuroscience},
  year = {2014},
  doi = {10.3389/fnhum.2014.00817},
}
Figure