Classification of Motor Imagery for Ear-EEG based Brain-Computer Interface

Brain-computer interface (BCI) researchers have shown an increased interest in the development of ear-electroencephalography (EEG), which is a method for measuring EEG signals in the ear or around the outer ear, to provide a more convenient BCI system to users. However, the ear-EEG studies have researched mostly targeting on a visual/auditory stimuli-based BCI system or a drowsiness detection system. To the best of our knowledge, there is no study on a motor-imagery (MI) detection system based on ear-EEG. MI is one of the mostly used paradigms in BCI because it does not need any external stimuli. MI that associated with ear-EEG could facilitate useful BCI applications in real-world. Hence, in this study, we aim to investigate a feasibility of the MI classification using ear-around EEG signals. We proposed a common spatial pattern (CSP)-based frequency-band optimization algorithm and compared it with three existing methods. The best classification results for two datasets are 71.8% and 68.07%, respectively, using the ear-around EEG signals (cf. 92.40% and 91.64% using motor-area EEG signals).