A comparison of classification performance among the various combinations of motor imagery tasks for brain-computer interface

Motor imagery brain-computer interface (BCI) is a system that sends commands from human to external devices using brain activity patterns of imagination of a motor action without an actual movement. In this paper, we compared classification performance among the various combinations of motor imagery tasks, toward the multi-dimensional control of motor imagery BCI. We used EEG motor imagery dataset of 99 subjects. Common spatial patterns (CSP) and linear discriminant analysis (LDA) were applied to extract features and to classify motor imagery tasks. 10×10 fold cross validation was used to evaluate classification accuracies through large dataset. For two-class discrimination, we compared the classification accuracy of the results between combinations: both feet and one hand, and both hand and one hand. From these results, using both feet motor imagery task showed 3% higher accuracy than using both hand motor imagery task (p<;0.01). For four-class discrimination, the compared result of classification between left/right/both hand/rest and left/right/both feet/rest showed that there was no significant difference between above combinations.