Enhancing Brain Machine Interface Decoding Accuracy through Domain Knowledge Integration

Area of Science:

• Neuroscience
• Biomedical Engineering
• Robotics

Background:

• Brain-machine interfaces (BMIs) decode neural signals for device control.
• Accurate estimation of muscle activity is crucial for effective BMI function.
• Current BMI decoding methods often lack integration of motor control principles.

Purpose of the Study:

• To introduce a novel decoding approach for BMIs that leverages domain knowledge of motor control.
• To enhance the accuracy and stability of muscle activity estimation in BMIs.
• To improve BMI performance by incorporating insights into the relationship between torque direction and muscle activity.

Main Methods:

• Developed a Kalman filter augmented with models of muscle activity and torque for specific movement directions.
• Integrated domain knowledge of motor control, specifically the relationship between torque direction and muscle activity.
• Validated the approach using decoding analysis with non-human primates performing an isometric wrist torque tracking task.

Main Results:

• Demonstrated significant improvements in muscle activity estimation accuracy compared to a standard Kalman filter.
• Showcased enhanced stability in muscle activity estimation.
• Validated the effectiveness of the domain knowledge integration in a real-world task.

Conclusions:

• The proposed decoding approach significantly enhances BMI performance by incorporating motor control domain knowledge.
• This method offers a promising direction for developing more accurate and stable BMIs.
• Leveraging domain-specific insights is key to advancing BMI technology.