High-performance brain-machine interface enabled by an adaptive optimal feedback-controlled point process decoder.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Kalman filters (KF) with closed-loop decoder adaptation (CLDA) have advanced brain-machine interface (BMI) performance.
• Recent decoders like ReFIT-KF and SmoothBatch-KF utilize CLDA for improved control.
• Existing methods adapt parameters over minutes and control at a slower rate.

Purpose of the Study:

• To introduce and evaluate a novel closed-loop BMI architecture: the adaptive optimal feedback-controlled (OFC) point process filter (PPF).
• To demonstrate high-performance and robust BMI control using the adaptive OFC-PPF system.
• To compare the adaptive OFC-PPF against established KF-based decoders.

Main Methods:

• Developed an adaptive OFC-PPF architecture for real-time BMI control.
• Enabled neural command issuance and feedback with every spike event.
• Implemented spike-by-spike adaptation of decoder parameters.
• Utilized an infinite-horizon OFC model for inferring velocity intention.
• Tested performance in a self-paced center-out movement task with 8 targets in a monkey.

Main Results:

• The adaptive OFC-PPF demonstrated improved BMI control compared to SmoothBatch-KF.
• Enhanced control and feedback rates were observed due to the PPF's spike-level processing.
• The OFC model provided a better approximation of user strategy, contributing to performance gains.
• Spike-by-spike adaptation led to faster convergence of BMI performance.
• Proficient BMI control was achieved in the monkey subject.

Conclusions:

• The adaptive OFC-PPF represents a significant advancement in BMI technology.
• This novel architecture offers superior performance through faster processing and adaptive capabilities.
• The OFC-PPF system holds promise for more intuitive and effective neural control applications.