Effects on the precision of a brain-computer interface when reducing the number of neural recordings used as input

Sammanfattning: A brain-computer interface (BCI) is a technology where electric activity from the brain is interpreted by a computer in order to perform a specific task. There exists various methods to record brain activity depending on if the recording devices are implanted under or over the scalp. BCIs often use a device called a microelectrode array (MEA) which is implanted directly into the brain. When implanting MEAs, different problems can arise such as brain tissue response, neural cell loss and damage to the user. Furthermore, the change in neural activity can also cause problems for BCI performance. This study aims to analyze how the performance of a BCI decoder is affected by a decrease in neural recordings used as input. Transfer learning was applied to an existing BCI which decodes neural recordings from attempted handwritten sentences that a paralysed user was imagining writing and translates it into text. The performance of the BCI decoder was evaluated using three different approaches to reducing neural recordings. The results showed that electrodes with high activity recorded had a significant impact on the decoder performance. The results also indicated that electrodes with low activity had insignificant im- pact on the performance. This suggests that electrode count may not be an important factor for the decoder performance and that fewer electrodes could be used and still achieve high BCI performance. To get a better understanding if this is possible, further research is needed. For instance, evaluating the performance when retraining the decoder with neural recordings from fewer electrodes could be performed.