Volitional control of movement interacts with proprioceptive feedback in motor cortex during brain-computer interface control in humans

Abstract

AbstractVision and proprioception regulate motor output during reaching. To study the effects of sensory input on motor control, brain computer interfaces (BCIs) offer particular advantages. As part of a long-term clinical BCI trial, we implanted two 96-channel microelectrode arrays into M1 of a person who was completely paralyzed below the neck but retained intact somatosensation. Neural recordings from M1 were transformed into a 2-dimensional velocity control signal for a robotic arm using an optimal linear estimator decoder that was calibrated while the participant imagined performing movements demonstrated by a virtual arm. Once the decoder was calibrated, we asked the participant to move the robotic arm left and right past a pair of lines as many times as possible in one minute. We examined how visual and proprioceptive feedback were incorporated into BCI control during this task by providing the participant with either visual or proprioceptive feedback, both, or neither. Proprioceptive feedback was provided by moving the participant’s own arm to match the movement of the robotic arm. Task performance with vision or proprioception alone was better than when neither were provided. However, providing proprioceptive feedback impaired performance relative to visual feedback alone, unless the decoder was calibrated with neural data collected while both visual and proprioceptive feedback were provided. Providing proprioceptive feedback during decoder calibration rescued performance because it better captured M1’s neural activity during BCI control with proprioceptive feedback. In general, BCI performance was positively correlated with how well the decoder captured variance in neural activity during the task. In summary, we found that while the BCI participant was able to use proprioceptive feedback regardless of whether the decoder was trained with vision only or vision and proprioception, training the decoder with both visual and proprioceptive feedback made performance more robust to the addition or removal of visual or proprioceptive feedback. This was because training a decoder with proprioceptive feedback allows the decoder to take advantage of proprioception-driven activity in M1. Overall, we demonstrated that natural sensation can be effectively combined with BCI to improve performance in humans.