Time-varying MVAR algorithms for directed connectivity analysis: Critical comparison in simulations and benchmark EEG...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Human brain function relies on rapid, directed interactions between brain regions.
• Quantifying dynamic functional connectivity requires high temporal resolution methods.
• Existing time-varying multivariate autoregressive (tvMVAR) models lack systematic comparative analysis.

Purpose of the Study:

• To critically compare the performance of four recursive tvMVAR algorithms.
• To assess the sensitivity of tvMVAR models to parameter choices (adaptation coefficients, model order, sampling rate).
• To compare single-trial vs. multi-trial modeling approaches for estimating dynamic connectivity.

Main Methods:

• Systematic comparison of four recursive tvMVAR algorithms.
• Performance evaluation using numerical simulations and benchmark EEG data.
• Varying key parameters: adaptation coefficients, model order, and signal sampling rate.
• Comparison of single-trial averaging and multi-trial modeling strategies.

Main Results:

• All tested tvMVAR algorithms accurately reproduced interaction patterns across various model orders.
• Signal downsampling generally reduced connectivity estimation accuracy, but could decrease variability.
• Single-trial modeling with larger adaptation coefficients outperformed previous recommendations and showed slower adaptation than multi-trial modeling.

Conclusions:

• Identified strengths and weaknesses of current tvMVAR approaches for analyzing dynamic brain connectivity.
• Provided practical recommendations for applying tvMVAR models to electrophysiological data.
• Highlighted the importance of parameter selection and modeling strategy for accurate connectivity estimation.