Benchmarking metrics for inferring functional connectivity from multi-channel EEG and MEG: A simulation study

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Inferring brain connectivity from electroencephalogram (EEG) and magnetoencephalogram (MEG) signals is crucial for understanding neural dynamics.
• Numerous metrics exist to quantify synchronization (magnitude, amplitude, phase), but their reliability in real-world neuroimaging data is often unclear.
• Spatial leakage and varying sensitivity to signal properties can compromise the validity of connectivity analyses.

Purpose of the Study:

• To systematically evaluate the reliability of commonly used EEG/MEG connectivity metrics.
• To compare metrics based on immunity to spatial leakage, test-retest reliability, and sensitivity to noise, coupling strength, and synchronization transitions.
• To guide the selection of appropriate connectivity metrics for EEG/MEG data analysis.

Main Methods:

• A biophysical model generating EEG/MEG-like signals was employed.
• A system of coupled chaotic oscillators simulated transitions between phase and amplitude synchronization.
• Five reliability benchmarks were used to assess coherence (Coh), imaginary coherence (ImCoh), amplitude envelope correlation (AEC), corrected AEC (AECc), phase coherence (PCoh), phase lag index (PLI), and weighted PLI (wPLI).

Main Results:

• Coh, AEC, and PCoh exhibited significant spatial leakage, rendering them unsuitable for real EEG/MEG data.
• ImCoh, AECc, PLI, and wPLI showed reduced spatial leakage, with PLI and wPLI demonstrating the highest immunity.
• PLI and wPLI outperformed ImCoh and AECc in test-retest reliability and sensitivity to coupling strength and synchronization transitions.
• AECc presented lower noise levels compared to ImCoh, PLI, and wPLI.

Conclusions:

• The reliability of EEG/MEG connectivity metrics is highly variable.
• Metrics like PLI and wPLI are recommended due to their robustness against spatial leakage and superior performance in capturing synchronization dynamics.
• The choice of connectivity metric must be carefully considered based on specific research questions and the assessed reliability factors.