Channels Selection for Motor Imagery Paradigm - an Itakura Distance based Method

An offline analysis method is proposed for a brain computer interface paradigm. Changes that appear in brain during the motor tasks should be reflected in the EEG signals. The sequences of EEG data are modeled by autoregressive (AR) processes. Based on Itakura distance (ID), the differences that occur during mental tasks (left and right hand movement imagination) versus relaxation period are measured. After applying statistical tests, channels selection is performed. The data contained in the chosen channels are classified with linear discriminant classifier (LDA), quadratic discriminant classifier (QDA) and Mahalanobis distance classifier (MD). The advantage of channels selection based on ID is that the picked channels contain relevant features. The effectiveness of the method is sustained by the classification rates obtained.