Modeling time-varying brain networks with a self-tuning optimized Kalman filter

Area of Science:

• Neuroscience
• Computational Neuroscience
• Dynamical Systems

Background:

• Brain networks exhibit complex, rapidly changing directed interactions crucial for cognitive and motor functions.
• Modeling these dynamic brain networks is challenging due to non-stationary neural signals and unknown noise.
• Accurate tracking of functional connectivity over time is essential for understanding brain function.

Purpose of the Study:

• To introduce a novel adaptive filter, the Self-Tuning Optimized Kalman filter (STOK), for modeling dynamic brain networks.
• To enhance the accuracy, temporal precision, and noise robustness of tracking directed functional connectivity.
• To validate the STOK filter's performance against traditional methods using simulations and real neural data.

Main Methods:

• Developed an innovative Kalman filter formulation with self-tuning memory decay and recursive regularization.
• Implemented the Self-Tuning Optimized Kalman filter (STOK) algorithm.
• Validated STOK by comparing its performance against the classical Kalman filter using simulated data and electroencephalography (EEG) recordings.

Main Results:

• The STOK filter demonstrated significantly superior performance in estimating time-frequency patterns of directed connectivity compared to the classical Kalman filter.
• STOK effectively recovered latent structures of dynamic connectivity from rat and human EEG data.
• Results showed excellent agreement between STOK-derived connectivity patterns and known neurophysiological principles.

Conclusions:

• The STOK filter is a powerful and robust tool for modeling dynamic network structures in biological systems.
• This algorithm offers improved accuracy and precision for tracking rapidly evolving brain network states.
• STOK has the potential to advance our understanding of the neural basis of brain functions.