Robust Bayesian Estimation of EEG-Based Brain Causality Networks.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biomedical Engineering

Background:

• Multivariate autoregression (MVAR) models are crucial for constructing brain causality networks.
• Outliers in EEG signals, like head movements, significantly impact MVAR parameter estimation accuracy, especially in short time windows.

Purpose of the Study:

• To propose a robust MVAR parameter estimation method resilient to outliers and effective in short time windows.
• To enhance the accuracy and reliability of brain network construction from EEG data.

Main Methods:

• Developed Lap-SBL, a Bayesian probabilistic framework using Laplace fitting error for MVAR parameter estimation.
• Modeled fitting error with a Laplace distribution to mitigate outlier influence, unlike traditional Gaussian models.
• Employed convex analysis and variational inference for efficient parameter estimation.

Main Results:

• Lap-SBL demonstrated reduced parameter estimation bias and more consistent network linkages compared to benchmark methods (LS, LASSO, LAPPS, SBL).
• Analysis of motor imagery data showed Lap-SBL effectively captures brain network lateralization characteristics.
• The method's performance highlights its capability in handling noisy EEG data.

Conclusions:

• Lap-SBL robustly suppresses outlier influence, enabling reliable brain network recovery even with short EEG windows.
• The proposed method offers a significant advancement for analyzing brain connectivity in challenging conditions.