Improved Neurophysiological Process Imaging Through Optimization of Kalman Filter Initial Conditions.

Area of Science:

• Neuroscience
• Biophysics
• Computational Neuroscience

Background:

• Electromagnetic source imaging techniques are crucial for understanding brain activity.
• Existing methods for neurophysiological process imaging often rely on Kalman filters for state and parameter inference.
• The performance of Analytic Kalman Filters (AKF) is highly sensitive to initialization, which is challenging due to unavailable ground truth data.

Purpose of the Study:

• To develop an efficient framework for optimizing AKF initialization in neurophysiological imaging.
• To address the limitations of conventional optimization techniques for AKF initialization.
• To improve the accuracy and efficiency of inferring neural mass model parameters for brain dynamics.

Main Methods:

• Implemented a novel blackbox optimization framework to find optimal AKF initializations.
• Compared multiple state-of-the-art optimization methods, focusing on Gaussian Process Optimization.
• Evaluated the framework using simulation data and real magnetoencephalography (MEG) data.

Main Results:

• Gaussian Process Optimization reduced the objective function by 82.1% and parameter estimation error by 62.5% on average in simulations.
• The framework achieved an average 13.2% reduction in the objective function on real MEG data.
• Demonstrated significant improvements in neurophysiological process imaging compared to unoptimized initialization.

Conclusions:

• The proposed blackbox optimization framework provides an efficient solution for AKF initialization.
• This method enhances the accuracy of inferring neural dynamics from electromagnetic source imaging.
• The optimized framework offers a valuable tool for uncovering complex underpinnings of brain activity.