An extended multivariate autoregressive framework for EEG-based information flow analysis of a brain network

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Effective connectivity studies are crucial for understanding brain function using high-resolution Electroencephalography (EEG) data.
• Traditional Granger Causality (GC) analysis, based on strictly causal multivariate autoregressive (MVAR) models, fails to account for instantaneous neural interactions.
• Ignoring instantaneous interactions in GC analysis can lead to inaccurate conclusions about brain information flow.

Purpose of the Study:

• To address the limitations of traditional GC analysis in capturing instantaneous neural interactions.
• To introduce an extended MVAR (eMVAR) model capable of incorporating zero-lag interactions.
• To propose and evaluate a novel constrained adaptive Kalman filter (CAKF) approach for eMVAR model identification.

Main Methods:

• Application of an extended multivariate autoregressive (eMVAR) model to EEG data.
• Development and implementation of a constrained adaptive Kalman filter (CAKF) for eMVAR model identification.
• Comparison of the CAKF approach with short time windowing-based adaptive estimation methods.

Main Results:

• The proposed CAKF approach effectively identifies eMVAR models, accounting for instantaneous interactions.
• The CAKF method demonstrates superior performance compared to short time windowing techniques in information flow analysis.
• Accurate analysis of brain information flow is achieved by considering zero-lag interactions.

Conclusions:

• The eMVAR model, combined with the CAKF approach, provides a more accurate method for analyzing brain information flow from EEG data.
• This method overcomes the limitations of traditional GC by incorporating instantaneous neural interactions.
• The CAKF-based eMVAR identification is a significant advancement for effective connectivity research.