Investigation of EEG signals generated by motor imagery for application in BCIs

Abstract

Motor imagery (MI) is a commonly used strategy in brain-computer interfaces (BCIs) to modify neuronal activity, in which the user, by imagining motor movements, generates signals that can be recorded and interpreted to control a device. In this study, we sought to investigate how the brain response of users during MI happens, by analyzing a database of EEG signals in which healthy subjects were asked to imagine the movement of their right and left hands. Our goal was to recognize patterns associated with this task, through a spectral evaluation of different segments of the signal. Therefore, we estimated the power spectral density (PSD) for each evaluated segment and then used it for classification, via k-nearest neighbors (k-NN). We found that the accuracy rates obtained with k-NN classification were very similar to random, suggesting, mainly, high inter-subjects variability and choice of a low complexity classifier.