Utilizing Subject-Specific Discriminative EEG Features for Classification of Motor Imagery Directions

Electroencephalogram (EEG)-based BrainComputer Interface (BCI) technology needs efficient algorithms to find distinct EEG patterns/features to realize applications with distinct high-dimensional control signals. This paper proposes a novel feature extraction methodology for separating EEG patterns associated right hand motor imagery performed towards left and right directions. The most discriminative subject-specific feature set is chosen based on Fisher's ratio of absolute phase values of EEG in 6 low frequency sub bands. Using this, the proposed BCI system is capable of providing better classification results than state-of-the-art methodology with fixed channels, fusing absolute phase and spatial features from selected subject-specific discriminative channels. Experimental analysis shows that though parietal lobe is vital in providing distinguishable features, the channel set that provide maximum accuracy, is highly subject-specific. Hence, subject-specific BCI that can decode finer parameters of imagined movement are feasible and further research to understand the activations elicited in parietal lobe can contribute towards robust BCI systems.