Selective Subject Pooling Strategy to Achieve Subject-Independent Motor Imagery BCI

Brain-computer interface (BCI) has facilitated communication for people who cannot move their bodies. BCI system requires time-consuming calibration phase to make reasonable classifier. To reduce the calibration phase, it is natural to attempt to make cross-subject classifier using other subjects' data. However, electroencephalogram (EEG) data are notably varied over subjects, that is, subject-specific. Thus, it is challenging to make subject-independent BCI performance comparable to subject-specific BCI performance. In this study, we investigated subject-independent motor imagery BCI performance with selective subjects (choosing subjects yielding reasonable performance selectively) instead of using all available subjects. We observed from MI-BCI dataset including 52 subjects that selective subject pooling strategy worked reasonably. Finally, criterion of selection of subjects for subject-independent BCI was suggested.