TL;DR
This paper introduces a new method, Anti-Symmetrized Bispectrum PAC (ASB-PAC), to reliably distinguish true brain-region communication (between-site PAC) from signals that appear coupled due to volume conduction or source leakage (spurious PAC). This is crucial for accurately interpreting brain activity and understanding how different brain areas coordinate.
What Problem Does This Paper Address?
Phase-amplitude coupling (PAC) is a phenomenon where the timing of a fast brain oscillation is locked to the timing of a slower oscillation. It's thought to be a key mechanism for integrating information across different brain areas. However, in real electrophysiological recordings like EEG, signals from different brain regions can mix together due to volume conduction (signals spreading through tissue) or imperfect source reconstruction. This mixing can create the appearance of PAC between brain regions even when the true coupling is only happening within a single region (within-site PAC). This makes it hard to tell if observed PAC reflects genuine communication between distinct brain sites or is just an artifact of signal mixing. The paper aims to solve this by developing a method that can reliably distinguish true between-site PAC from these spurious signals.
Methods
The core of the paper's contribution is the Anti-Symmetrized Bispectrum PAC (ASB-PAC) method. Traditional PAC measures, like the Modulation Index (MI) and standard bispectrum, are vulnerable to signal mixing. The authors leverage the property that anti-symmetrized bispectra should ideally be zero for mixtures of independent sources. By applying anti-symmetrization to the bispectrum, they create a PAC metric that is less susceptible to spurious signals arising from volume conduction.
They evaluated ASB-PAC against other methods (MI, MI with orthogonalization, ICA-based surrogates, and standard bispectrum) using:
- Simulations: They created controlled scenarios with varying levels of signal mixing, signal-to-noise ratios (SNR), and numbers of interacting sources. These simulations ranged from simple two-channel setups to more complex whole-brain EEG models.
- Real EEG Data: They applied ASB-PAC to EEG data from a motor imagery experiment.
- Software Update: They also present an updated version of the free ROIconnect plugin for EEGLAB, which now includes ASB-PAC functionality.
Datasets
The paper primarily uses simulated datasets for evaluation. These simulations were designed to mimic realistic EEG recordings, including:
- Two-channel experiments: Simulating both true bivariate (between-site) PAC and univariate (within-site) PAC, with and without signal mixing.
- Whole-brain EEG experiments: Using realistic EEG forward models, source reconstruction, and dimensionality reduction techniques to simulate complex scenarios with varying SNRs and numbers of interactions.
- Real EEG data: Data from a motor imagery experiment (VitalBCI study) was used for exploratory analysis.
Key Insights and Ideas
- Signal mixing is a major confound for PAC: The paper strongly demonstrates how volume conduction and source leakage can create convincing but false between-site PAC signals.
- ASB-PAC is remarkably robust to signal mixing: The anti-symmetrization technique effectively suppresses spurious PAC, allowing for a clearer view of genuine inter-regional communication.
- Orthogonalization and ICA surrogates are not sufficient for PAC: While useful for other connectivity measures, these general methods don't adequately address the specific challenges of distinguishing between-site PAC from within-site PAC artifacts.
- Bispectral methods are generally better than MI for PAC detection: Especially in the presence of volume conduction, bispectral approaches, and particularly ASB-PAC, outperform traditional MI-based metrics.
Key Results
- ASB-PAC outperforms other methods: In simulations, ASB-PAC consistently showed the highest performance in detecting genuine between-site PAC while exhibiting the lowest false-positive rates, especially in the presence of signal mixing.
- High True Negative Rate (TNR) for ASB-PAC: Even in challenging scenarios with significant signal mixing, ASB-PAC maintained a high TNR (at least 0.96 across SNRs in two-channel experiments), meaning it rarely falsely identified within-site PAC as between-site PAC.
- Bispectrum is a good starting point: The standard bispectrum also showed better performance than MI in detecting PAC in volume-conduction settings, but ASB-PAC further improved upon this by specifically addressing mixing artifacts.
- Real data application: Exploratory analysis on motor imagery EEG data demonstrated the utility of ASB-PAC in identifying potential between-region PAC.
Limitations and Future Work
- Potential for spurious PAC from other sources: The authors acknowledge that spurious PAC can arise from factors other than volume conduction, such as physiological artifacts (eye movements, muscle activity), non-sinusoidal signals and their harmonics, and other types of cross-frequency coupling (phase-frequency, amplitude-frequency). They suggest these possibilities need further investigation.
- Harmoni method: A novel method called Harmoni has been developed to address spurious PAC due to higher harmonics. Future research could explore combining Harmoni with ASB-PAC.
- Ambiguity in genuine between-site PAC: In cases where a univariate PAC signal is reflected in two channels with different time delays, ASB-PAC might still be ambiguous. The authors suggest that testing for stable phase delays between channels could help resolve this.
- Computational cost: While not explicitly detailed as a limitation, bispectral methods are generally more computationally intensive than MI.
- Further validation of ASB-PAC interpretation: While ASB-PAC shows excellent performance in distinguishing signal types, further work could solidify its physiological interpretation in complex brain states.
Summary updated