Phase Synchronization Between EEG Signals as a Function of Differences Between Stimuli Characteristics

L. ten Bosch, K. Mulder, L. BovesInterspeech 2019

TL;DR

This paper introduces a novel method for analyzing EEG signals that combines traditional ERP analysis with measures of phase synchronization and power in specific frequency bands. By applying this to an auditory lexical decision task, the authors demonstrate that phase synchronization patterns can effectively distinguish between different types of speech stimuli, offering a richer understanding of neural processing beyond amplitude-based ERPs.

What Problem Does This Paper Address?

Understanding how the brain processes speech in real-time is a complex challenge. While Event-Related Potentials (ERPs), derived from EEG, have been instrumental in revealing neural responses to stimuli, they primarily focus on signal amplitude over time. This approach often overlooks the crucial role of phase synchronization, which describes how different neural oscillations align in time. Existing ERP analysis can be dominated by low-frequency components, obscuring finer details related to cognitive processes like semantic or acoustic processing. This paper aims to bridge this gap by developing a unified framework to analyze both ERPs and phase synchronization, allowing for a more comprehensive investigation of how neural activity reflects differences in speech stimuli.

Methods

The authors propose a signal processing pipeline that allows for the comparison of ERPs, instantaneous power, and phase synchronization. For phase synchronization, they employ two main approaches:

  • Analytic Signal and Hilbert Transform: This method involves band-pass filtering the EEG signal into specific frequency bands (theta: 4-8 Hz, alpha: 8-12 Hz) and then using the Hilbert transform to extract the analytic signal. The phase of this analytic signal provides the instantaneous phase.
  • Wavelet Transform: Complex Morlet wavelets are used to decompose the signal into different frequency components, allowing for the extraction of both amplitude and phase.

To quantify phase synchronization between EEG signals, they utilize the Phase Locking Value (PLV). PLV measures the consistency of the phase difference between two signals over time. To make this interpretable, they average signals from multiple participants and conditions, then compute circular variance on the resulting relative phase patterns. Circular variance is a measure of dispersion for circular data, and a decrease in circular variance indicates increased phase synchronization. The paper also details methods for computing instantaneous power using the Hilbert transform.

Datasets

The study uses EEG data collected from 29 participants performing an auditory lexical decision task on English words. The stimuli were designed with three nested factors:

  1. Syllable Reduction: Presence/absence of a reduced syllable.
  2. Syllable Position: Position of the reduced syllable relative to the stressed syllable (pre-stress or post-stress).
  3. Cognate Status: Whether the target word was a cognate (similar meaning and form in Dutch and English) or a control (non-cognate).

This resulted in eight distinct sets of target stimuli, allowing for the investigation of how phase synchronization differs across these conditions.

Key Insights and Ideas

  • Phase Synchronization Reflects Stimulus Differences: The study demonstrates that phase synchronization patterns in EEG signals are sensitive to variations in speech stimuli, particularly in the theta band.
  • Unified Analysis Framework: The proposed method allows for the simultaneous analysis of ERPs, power, and phase synchronization, offering a more holistic view of neural processing.
  • Circular Variance as a Key Metric: Circular variance effectively quantifies the degree of phase synchronization, showing a clear relationship with stimulus contrasts. A lower circular variance (higher PLV) indicates stronger phase synchronization.
  • Temporal Dynamics of Synchronization: Phase synchronization emerges around 100 ms after word onset and fades around 400 ms, suggesting specific temporal windows for these neural processes.
  • Complementary Information: Both power and phase synchronization provide valuable, albeit different, information about neural responses to auditory stimuli.

Key Results

  • The regression analysis showed that stimulus contrasts (stress, cognate status, and reduction) significantly predicted circular variance, indicating differences in phase synchronization.
  • Specifically, the circular variance was significantly higher (meaning lower phase synchronization) for contrasts like prestress vs. poststress, and full vs. reduced stimuli, especially in the alpha band compared to the theta band.
  • The PLV analysis revealed a "bridging" effect in relative phase plots around 100-400 ms, indicative of phase synchronization between groups of EEG signals.
  • The power analysis indicated that instantaneous power also varied significantly with frequency band and electrode site, and was influenced by stimulus factors.

Limitations and Future Work

  • Interpretability of Individual Signals: The authors note that the phase of individual EEG signals is difficult to interpret, necessitating averaging across participants and conditions.
  • Computational Complexity: Computing phase synchronization measures is more complex than traditional ERP analysis.
  • Future Directions: The authors suggest further research into phase effects using wavelet transforms, which might offer more nuanced insights. They also emphasize the need for transparent and accessible software tools to facilitate wider adoption of these advanced analysis techniques. The current work lays the groundwork for future experiments to explore these interactions in greater detail.

Summary updated

BibTeX
@inproceedings{bosch2019phase,
  author = {L. ten Bosch and K. Mulder and L. Boves},
  title = {{Phase Synchronization Between EEG Signals as a Function of Differences Between Stimuli Characteristics}},
  booktitle = {Interspeech},
  year = {2019},
}
Figure