Exploring connectivity of resting-state EEG between BCI-literate and -illiterate groups

Abstract

AbstractAlthough Motor Imagery-based Brain-Computer Interface (MI-BCI) holds significant potential, its practical application faces challenges attributable to the phenomenon known as BCI-illiteracy. BCI researchers have attempted to predict BCI-illiteracy to mitigate this issue. As connectivity’s significance in neuroscience has grown, BCI researchers have applied connectivity to predict BCI-illiteracy with resting-state data. However, connectivity metrics’ use and interpretation of the results can be challenging for several reasons. Firstly, there are various connectivity metrics, each with its own advantages and disadvantages based on their underlying hypotheses and perspectives. These pros and cons are shaped by several factors, increasing the complexity of their application and interpretation. Secondly, it is unclear whether they are as acceptable as their developers have claimed. Thirdly, it is not evident which factor may influence the estimation of connectivity and which metric is suitable for this research. Therefore, this study conducted an empirical test to provide BCI researchers with a better understanding of connectivity. We analyzed three large public datasets using three functional connectivity (FC) and three effective connectivity (EC) metrics. Additionally, the structural difference in the resting-state network between BCI-literate and illiterate groups was examined. Our analysis revealed that the appropriate frequency range to measure connectivity varies depending upon the metric used. The alpha range was found to be suitable for FC, while the alpha, alpha + theta, and beta ranges were found to be appropriate for EC. Further, the results of estimating connectivity varied depending upon the dataset and metric used. Although we observed that BCI-literacy had stronger connections between nodes, no other significant structural differences were found between the two groups. However, BCI-literacy’s resting-state network displayed higher network efficiency compared to BCI-illiteracy, regardless of the metrics and dataset used. Therefore, it seems reasonable to use resting-state connectivity to predict BCI-illiteracy. Our conclusion is that each metric has its own specific hypothesis and perspective to measure connectivity under certain conditions.