Novel Approach to Classify Motor-Imagery EEG Recordings with Convolutional Neural Network Using Network Measures

Electroencephalogram (EEG) signal recorded during motor imagery (MI) tasks has been widely applied in brain-computer interface (BCI) applications as communication approach. To improve the classification success rate of MI tasks, this paper proposes novel input form based on brain network connectivity measures to perform classification for MI EEG for the datasets from BCI Competition IV. Firstly, using connectivity patterns between brain regions during MI six more frequent network features were selected and their maps in form of 2-D images were generated; then simple yet powerful convolutional neural network (CNN) with convolutional layer was applied to perform binary classification of MI tasks (left-hand, right-hand, both feet and tongue movements). The discrimination ability of these features was compared with each other. Our results demonstrate that CNN fed with path length feature map can further improve classification performance in most binary problems. Whileall classification results are better than 86%, the best accuracy using brain network features is 96.69% in right-tongue separation. The present study shows that the proposed method is effective to classify MI, and provides practical method for the classification of non-invasive EEG signals in BCI applications.