Robust Brain-Machine Interface Design Using Optimal Feedback Control Modeling and Adaptive Point Process Filtering.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Control Theory

Background:

• Brain-machine interfaces (BMI) commonly use decoders like Kalman filters with closed-loop decoder adaptation (CLDA).
• Current decoders lack direct spike modeling, potentially limiting BMI control and adaptation speed.
• A unified CLDA framework for intention estimation, assisted training, and adaptation is needed.

Purpose of the Study:

• Develop a novel closed-loop BMI training architecture for processing, control, and adaptation using spike events.
• Establish a unified, control-theoretic CLDA framework for intention estimation, assisted training, and adaptation.
• Enhance BMI robustness, speed, and generalizability across tasks.

Main Methods:

• Incorporated an infinite-horizon optimal feedback-control (OFC) model for brain behavior in BMI.
• Utilized a point process model for spike events, enabling faster, spike-event-based processing and adaptation.
• Developed unified CLDA techniques for intention estimation and autonomous assisted training.

Main Results:

• OFC intention estimation enhanced BMI performance compared to existing methods.
• OFC assisted training facilitated consistent and proficient control.
• Spike-event-based adaptation demonstrated faster, more consistent performance convergence than batch-based methods.
• The architecture proved robust to parameter initialization and extended control to new tasks.

Conclusions:

• The novel BMI architecture offers significant improvements in processing speed, control, and adaptation.
• The unified CLDA framework provides a systematic approach to intention estimation and assisted training.
• Results suggest implications for developing clinically viable neuroprosthetics.