Brain Connectivity Estimation Pitfall in Multiple Trials of Electroencephalography Data.

Area of Science:

• Neuroscience
• Signal Processing
• Biomedical Engineering

Background:

• Electroencephalography (EEG) offers high temporal resolution ideal for brain connectivity analysis.
• Calculating connectivity from multi-trial EEG data presents challenges regarding trial inclusion and averaging.
• Existing methods lack clarity on optimal trial handling for accurate connectivity estimation.

Purpose of the Study:

• To investigate the impact of trial number on brain connectivity estimation using EEG.
• To determine the most reliable method for computing connectivity from multi-trial EEG data.
• To provide guidelines for accurate brain connectivity analysis in neuroscience research.

Main Methods:

• Simulated EEG data with known auto-regressive (AR) coefficients were generated.
• Multivariate autoregressive (MVAR) models were extracted using varying numbers of trials.
• Granger causality (GC) and Partial Directed Coherence (PDC) were employed to estimate connectivity.

Main Results:

• Connectivity estimates were significantly higher and more variable with small trial numbers (5-10 trials).
• Increasing the number of trials improved MVAR model fit and led to converged connectivity values.
• Findings from simulated data were corroborated by analyses on real EEG data.

Conclusions:

• Brain connectivity should be calculated for each trial individually, followed by averaging the results.
• Larger trial numbers enhance MVAR model appropriateness and the reliability of connectivity estimations.
• Connectivity estimations based on insufficient trial numbers are considered invalid and misleading.