Automated Sleep Stage Scoring Using Brain Effective Connectivity and EEG Signals

Sleep staging is necessary for the diagnosis of sleep disorders and evaluating the quality of sleep. Scoring of sleep stages is mainly done manually by a specialist based on Polysomnography data and mainly EEG, which is very time consuming and costly. Hence, it is essential to provide an automated method. This paper proposes an automatic sleep staging method based on brain effective connectivity. In this study, using the Granger causality criterion, causality matrices for each epoch of EEG data sampled from 10 healthy individuals were extracted as features. Then, the Gaussian SVM classifier has been employed to classify sleep stages using extracted features. For feature reduction, two algorithms, PCA and RSFS, were assessed, but we did not apply feature reduction in the final method due to the insignificant effect on classification accuracy. Finally, we were able to classify sleep stages with 72.7% accuracy and Cohen's Kappa Coefficient of 0.65. The experimental results demonstrate that the combination of Granger causality features and SVM can be used as an efficient framework for automated sleep stage scoring with regard to promising classification performance in terms of accuracy and Cohen's Kappa coefficient.