Closed-loop decoder adaptation on intermediate time-scales facilitates rapid BMI performance improvements independent...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Rehabilitation Technology

Background:

• Closed-loop decoder adaptation (CLDA) is vital for advancing brain-machine interface (BMI) performance.
• Rapid adaptation is crucial for clinical BMI applications, especially for patients with motor deficits and limited movement.
• The timescale of decoder adaptation is critical, particularly when initial BMI performance is low.

Purpose of the Study:

• To introduce SmoothBatch, a CLDA algorithm designed for rapid performance improvement.
• To evaluate SmoothBatch's effectiveness across different seeding methods with varying offline decoding power.
• To investigate the co-adaptation process between subjects and decoders in closed-loop BMIs.

Main Methods:

• Developed SmoothBatch, a CLDA algorithm updating decoder parameters on a 1-2 minute timescale using an exponentially weighted sliding average.
• Tested SmoothBatch in a nonhuman primate performing a center-out reaching BMI task.
• Seeded the algorithm with four distinct conditions: visual cursor observation, ipsilateral arm movements, baseline neural activity, and arbitrary weights.

Main Results:

• SmoothBatch demonstrated rapid performance enhancement, increasing success rates from 0.018 to over 8 successes/min within approximately 13 minutes, irrespective of the seeding method.
• Sustained high performance was observed even after the adaptation phase concluded.
• Decoder adaptation convergence paralleled performance improvements, indicating a co-adaptation dynamic between the subject and the decoder.

Conclusions:

• SmoothBatch effectively and rapidly improves closed-loop BMI performance, even with limited initial decoding power.
• The algorithm's success across various seeding strategies highlights its robustness for clinical translation.
• The study suggests that CLDA involves a synergistic co-adaptation process between the user and the BMI system.