Augmenting intracortical brain-machine interface with neurally driven error detectors.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Rehabilitation Technology

Background:

• Brain-machine interfaces (BMIs) currently require users to manually correct errors, which is time-consuming and inefficient.
• Identifying neural correlates of error perception could enable automated error correction in BMIs.
• The presence of intracortical outcome error signals in premotor and primary motor cortices for BMIs was previously unconfirmed.

Purpose of the Study:

• To investigate the presence of neural outcome error signals in the premotor and primary motor cortices during intracortical BMI task performance.
• To develop and implement a novel 'detect-and-act' system for automatic error correction in BMIs.

Main Methods:

• Recorded spiking activity in rhesus macaques performing an intracortical BMI computer cursor task.
• Developed a real-time system to decode BMI trial outcomes from neural activity.
• Implemented an 'error detect-and-act' system operating in parallel to a kinematic BMI decoder.

Main Results:

• A putative outcome error signal was identified in spiking activity within the premotor and primary motor cortices.
• BMI trial outcomes were decoded with high accuracy (96% shortly after, 84% before trial end).
• The 'detect-and-act' system significantly improved BMI performance in a task with substantial errors, outperforming Kalman filter variants.

Conclusions:

• Neural error signals can be detected in brain regions commonly used for BMI control.
• An automated 'detect-and-act' BMI system can effectively correct errors in real-time.
• This approach holds significant promise for enhancing the clinical viability of BMIs for motor function restoration.