A single-trial multi-class classification of various motor imagery tasks for EEG-based brain-computer interface communication

We studied the brain activity (alpha and beta rhythms) with various motor imagery tasks for improvement of BCI usability using 14 EEG electrodes in five healthy subjects. For this purpose, we estimated two-class and four-class classification accuracy on the EEG signals with four motor imagery tasks derived from each type motor imagery (three classical motor imagery and one proposed mental strategy) tasks using t-test and SVM. The proposed mental strategy was imagery writing Kanji (Japanese characters). It has the possibility of both sensorimotor cortex and the visual cortex activation. Therefore, we expected to extract the distinct activity different from the activation with classical motor imagery tasks. In the two-class classification results, the classification accuracy was 73.7% on average in all combination of derived motor imagery task. Moreover, we demonstrated that four-class classification accuracy was 40.1% and the proposed task had possibility of the visual cortex activation dominantly. In experimental results, we proposed the new way for improvement of BCI application usability.