Classification of unilateral motor tasks using spectral features of EEG

The research to identify the relationship between neural activity underlying motor tasks such as motor imagery, execution, planning and observation has been of significant interest, since it is an important aspect in improving our understanding of neuromotor networks. Neuroimaging literature suggests that the central mechanism in all these tasks is the activation of the primary motor areas of the brain. To study correlates of these tasks, the researchers often employ neural data recorded from the subjects as they perform motor tasks using their dominant hand. In this study, we investigate the distinction between EEG spectral features during unilateral right and unilateral left, imagined and executed hand motor tasks. The analysis is conducted for the data from each hand separately, with the objective to identify whether the EEG activity that distinguished unilateral motor tasks is distinct for each hand. Further, a novel approach of selection of multichannel EEG spectral features is proposed and supervised classification based on support vector machines is employed to classify unilateral motor imagination and motor execution tasks using the selected features. Grand averages of spectral features, over 10 subjects, are studied and it is observed that EEG activity during unilateral motor tasks are both spatially localized and lateralized. Using the proposed pattern recognition model, accuracies of 82.4 % and 83.1 % with a reduction in 15-20 % features for right and left unilateral motor tasks respectively.