Predicting c-VEP-based BCI performance to study BCI illiteracy
Keywords
Loading...
Authors
Issue Date
2023-01-27
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
A brain–computer interface (BCI) is a system that can deduce the intent of a user from recordings
of their brain activity. This allows users to control a computer application by brain activity,
which can be measured, e.g., by electroencephalography (EEG). After approximately 50 years of
BCI research, the success that BCI control can provide still greatly varies from subject-to-subject.
About 15% to 30% of sensorimotor rhythm-based (SMR) BCI users do not reach the criterion level
of 70% accuracy, which was determined to be the threshold for meaningful spelling. This is in the
literature known as the BCI illiteracy phenomenon. A myriad of variables are studied alongside
of subjects testing BCIs in order to determine if these variables correlate with BCI performance.
If they do, then they could potentially be used as predictors. The development of predictors of
BCI performance serves two purposes: it may lead to a better understanding of the BCI-illiteracy
phenomenon and these predictors could be used as a screening tool to inform users about poor
expected performance, among other things. An experiment was conducted in which six predictors
were analysed by means of a linear and a multivariate regression analysis alongside of the state-ofthe-
art paradigm code-modulated visual evoked potential-based (c-VEP) BCI performance. The
six predictors were: relative visual alpha power during resting state with eyes open and closed,
heart rate variability, attention span measured by the error rate of the Sustained Attention to
Response Task (SART) paradigm and flash-VEP latency and amplitude. There were no significant
(p > 0.008) Pearson’s correlation coefficients found for these predictors with N=16 and all subjects
were able to obtain sufficient BCI control when they were given enough time. It was concluded that
subject-to-subject variance is not in accuracy for c-VEP-based BCIs, but in time. The multivariate
regression model could be used as a screening tool given its root mean square error (RMSE) of
14.084% for a trial size of 1.05 seconds.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen