Round cosine transform based feature extraction of motor imagery EEG signals

Abstract

Brain Computer Interfaces (BCIs) are systems with great potential for the rehabilitation of people with severe motor injuries. By analyzing a subject’s brain waves, it is possible to detect patterns and translate his “thinking” into device commands, like prosthesis or a robotic arm. This research presents an EEG processing method, which is capable of detecting patterns of the subject’s motor imagery, splitting the patters in left or right hand imagery. The proposed method considers the Round Cosine Transform (RCT), a low computational complexity transform, and an artificial neural network (ANN) module which identifies the patterns. The method has been tested in a real-time (RT) continuous EEG processing experiment simulation, controlling a mouse arrow horizontally on a screen based on the subject’s imagery motor activity. The performance of the proposed method is evaluated in terms of the mutual information (MI), classification time and misclassification rate (%). The achieved results were 0.49 bits, 5.25 s and 15.6%, respectively.