Time-varying effective EEG source connectivity: the optimization of model parameters^.

Area of Science:

• Neuroscience
• Signal Processing
• Computational Biology

Background:

• Adaptive estimation methods, particularly the general Kalman filter, are crucial for studying dynamic brain networks due to the non-stationary nature of neural signals.
• Accurate selection of model order (p) and adaptation constant (c) is vital for reliable time-varying multivariate autoregressive (TV-MVAR) estimates, preventing frequency domain biases and temporal distortions.

Purpose of the Study:

• To establish an objective criterion for optimally selecting the model order (p) and adaptation constant (c) in adaptive Kalman filter-based brain network analysis.
• To address the limitations of residual- and information-based criteria that may not guarantee absolute minima for parameter selection.

Main Methods:

• Proposed a novel method utilizing partial derivatives of residual- and information-based criteria to guide the optimal selection of parameters p and c.
• Validated the proposed method using human visual evoked potentials (VEPs) during face perception, a well-understood neural process.
• Further validated the method with simulated data where the ground truth was known to assess performance rigorously.

Main Results:

• The developed objective criterion effectively guided the selection of optimal parameters (p and c) for adaptive Kalman filtering.
• Demonstrated improved accuracy and reduced biases in time-varying multivariate autoregressive (TV-MVAR) estimates compared to sub-optimal filtering.
• Validation with both real (human VEPs) and simulated data confirmed the method's efficacy in accurately capturing brain network dynamics.

Conclusions:

• The proposed partial derivative-based criterion offers a robust and objective approach for optimizing adaptive Kalman filter parameters in neuroscience.
• Accurate parameter selection is essential for reliable brain network dynamic analysis, leading to more trustworthy interpretations of neural signal propagation and information processing.
• This method enhances the precision of brain connectivity studies, contributing to a better understanding of neural processes like face perception.