Clustering of Preferred Directions During Brain-Computer Interface Usage

Brain-computer interfaces (BCIs) are proving to be viable clinical interventions for sufferers of amyotrophic lateral sclerosis, amputations, and spinal cord injuries. To improve the viability of BCIs, it will help to have a thorough understanding of how the brain controls them. Neural activity during usage of certain BCIs behaves in a surprising and seemingly counterintuitive manner – the preferred directions (PDs) of neurons cluster together. We trained monkeys to reach to targets in a center-out task either using their arm or a BCI. We found that neurons’ PDs cluster similarly during training of the BCI decoder and usage of the BCI, but remain relatively unclustered when the monkeys use their arms. Modulation depths increase upon usage of the BCI, and narrowness of tuning tends to either increase or decrease rather than staying the same. In addition, the cluster direction can be predicted from per-target performance. A model where two neurons’ PDs approach one another reveals how much modulation depths have to increase to maintain controllability. This thesis concludes with considerations of why this clustering might occur, and whether or not it benefits BCI control