Investigating causality between interacting brain areas with multivariate autoregressive models of MEG sensor data

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biophysics

Background:

• Estimating causal interactions between brain regions is crucial for understanding brain function.
• Magnetoencephalography (MEG) offers high temporal resolution for brain activity monitoring.
• Existing methods for causality analysis in MEG data face challenges with model robustness and computational time.

Purpose of the Study:

• To investigate the feasibility of estimating causal brain interactions using multivariate autoregressive (MAR) models fitted to MEG sensor data.
• To develop an efficient method for deriving whole-brain causality maps without prior region selection.

Main Methods:

• Fitting MAR models to MEG sensor-level data.
• Projecting sensor-level MAR model coefficients to source space to estimate causal interactions.
• Utilizing partial directed coherence (PDC) and MAR model coefficients for causality assessment.

Main Results:

• Demonstrated theoretical feasibility of estimating source-level causal interactions from sensor-level MEG data.
• Successfully reconstructed causal interactions in simulated MEG data when source locations were known.
• Showed that MAR model coefficients alone provide meaningful causality information in high-dimensional source spaces.
• The proposed method overcomes limitations of existing approaches, offering improved robustness and reduced computation time.

Conclusions:

• The developed methodology enables efficient derivation of entire brain causality maps from MEG data.
• This approach avoids the need for a priori selection of regions of interest, facilitating comprehensive brain network analysis.
• The findings support the use of sensor-level MAR modeling followed by coefficient projection for robust causal interaction estimation in neuroscience.