Identifying error features in a MI-BCI system using microstates

Abstract

Motor-Imagery Brain-Computer Interface (MI-BCI) is a useful method for identifying a user's motor intention when they imagine a certain movement. However, the accuracy of this approach is low, and varies depending on the performance of users. In order to overcome the limitation, we focused upon the detection of error due to the low MI-BCI performance. We used microstate analysis to produce a sparse feature set of large-scale brain network activity. Using this analysis, we developed a new feature from the error component. We conducted experiments with the 10 subjects (6 male, 4 female, aged 26.6 +/- 2.91 years) by presenting visual stimulus in real time. In the experiment, 30 % of all trials were configured to randomly generate error responses. The top five microstates to show the best correct/false responses were calculated. Microstates MS1 to MS4 produced the same results for both correct and incorrect responses, but microstate MS5 showed a difference between correct and false responses. Microstate MS5 is a representation of EEG response occurred within 400 ms after visual stimulus. We would say that `correct' and 'false' responses can be distinguished using the microstate MS5.