Regularized Kalman filter for brain-computer interfaces using local field potential signals.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Brain-computer interfaces (BCIs) are crucial for applications like prosthetic control, requiring accurate movement decoding from neural signals.
• The Kalman filter (KF) is commonly used for decoding in BCIs, but its performance depends on accurate estimation of system parameters.
• Traditional methods for estimating KF parameters can be suboptimal, limiting decoding accuracy in BCI applications.

Purpose of the Study:

• To enhance the decoding performance of Kalman filters in brain-computer interface systems.
• To improve the estimation of the state transition matrix and measurement noise covariance matrix for KF.
• To develop a more robust parameter estimation technique for BCI applications.

Main Methods:

• Proposed the Regularized Kalman Filter (RKF) incorporating two key features.
• Implemented regularization for the state equation regression to enhance state transition matrix estimation.
• Utilized a shrinkage method to improve the estimation of the unknown measurement noise covariance matrix.

Main Results:

• The Regularized Kalman Filter (RKF) demonstrated superior performance compared to conventional KF and other methods.
• Validation was performed using local field potential datasets from motor cortex recordings in monkeys and rats.
• The RKF achieved better decoding accuracy for kinematic and kinetic parameters during hand movements and force application.

Conclusions:

• The Regularized Kalman Filter (RKF) offers a significant improvement in decoding accuracy for BCI systems.
• The proposed method effectively addresses limitations in parameter estimation for Kalman filters.
• RKF provides a more robust and accurate approach for neural decoding in brain-computer interfaces.