Training the cortex to control three-dimensional movements of a neural prosthesis

In this study, rhesus macaques received implants in the motor and premotor areas of the brain. The implants were multichannel microwire electrode arrays which recorded the activity of up to 65 cortical units. This cortical activity was used in real time to control the movement of a cursor in a three-dimensional virtual workspace.; In the first experiment, two subjects made center-out movements of the virtual cursor with the cursor position being controlled either by their hand position (“hand-control”) or by recorded neural activity (“brain-control”). A modified population vector was used to translate the neural activity into cursor movements in real time under brain-control. In this “closed-loop” situation, the subjects had real-time visual feedback of the brain-controlled cursor. “Open-loop” trajectories were created offline by applying the same population vector algorithm to the neural activity recorded when the cursor was under hand-control. Closed-loop trajectories hit the targets significantly more often than the open-loop trajectories. Practice significantly improved the subjects' closed-loop performance both within and across days.; Two macaques also made brain-controlled cursor movements with both arms restrained. The directional tuning of the recorded units changed significantly under arms-restrained brain-control compared to the directional tuning seen during arms-free hand-controlled movements. A coadaptive algorithm was used to decode these new cortical modulation patterns into cursor movements. With regular practice, the units became more cosine tuned, increased their modulation range and reduced their movement-to-movement variability. This coadaptive algorithm allowed subjects to make target-directed brain-controlled movements of an accuracy level never achieved before using traditional cortical decoding methods.; The coadaptive brain-control algorithm was then tested in more practical applications. Subjects were able to make continuous sequences of movements to novel and trained target directions without further adaptation of the algorithm parameters. Movement accuracy improved with regular practice.; The coadaptive technique was also applied to direct brain-control of a robotic The subject was able to make center-out brain-controlled movements of the robot with nearly the same accuracy as with the virtual cursor. The subject also used the robot to make long continuous sequences of movements to random target positions without further adaptation of the algorithm parameters.