Detrended Fluctuation Analysis Features for Automated Sleep Staging of Sleep EEG

Abstract

An accurate sleep staging is crucial for the treatment of sleep disorders. Recently some studies demonstrated that the long range correlations of many physiological signals measured during sleep show some variations during the different sleep stages. In this study, detrended fluctuation analysis (DFA) is used to study the electroencephalogram (EEG) signal autocorrelation during different sleep stages. A classification of these stages is then made by introducing the calculated DFA power law exponents to a K-Nearest Neighbor classifier. The authors’ study reveals that a 2-D feature space composed of the DFA power law exponents of both the filtered THETA and BETA brain waves resulted in a classification accuracy of 93.52%, 93.52%, and 92.59% for the wake, non-rapid eye movement and rapid eye movement stages, respectively. The overall accuracy of the proposed system is 93.21%. The authors conclude that it might be possible to build an automated sleep assessment system based on DFA analysis of the sleep EEG signal.