Motor imagery task classification for brain computer interface applications using spatiotemporal principle component analysis

Classification of single-trial imagined left- and right-hand movements recorded through scalp EEG are explored in this study. Classical event-related desynchronization/synchronization (ERD/ERS) calculation approach was utilized to extract ERD features from the raw scalp EEG signal. Principle Component Analysis (PCA) was used for feature extraction and applied on spatial, as well as temporal dimensions in two consecutive steps. A Support Vector Machine (SVM) classifier using a linear decision function was used to classify each trial as either left or right. The present approach has yielded good classification results and promises to have potential for further refinement for increased accuracy as well as application in online brain computer interface (BCI).