A Self-adaptive Threshold Method for Automatic Sleep Stage Classification Using EOG and EMG

Abstract

Sleep stages are generally divided into three stages including Wake, REM and NRME. The standard sleep monitoring technology is Polysomnography (PSG). The inconvenience for PSG monitoring limits the usage of PSG in some areas. In this study, we developed a new method to classify sleep stage using electrooculogram (EOG) and electromyography (EMG) automatically. We extracted right and left EOG features and EMG feature in time domain, and classified them into strong, weak and none types through calculating self-adaptive threshold. Combination of the time features of EOG and EMG signals, we classified sleep stages into Wake, REM and NREM stages. The time domain features utilized in the method were Integrate Value, variance and energy. The experiment of 30 datasets showed a satisfactory result with the accuracy of 82.93% for Wake, NREM and REM stages classification, and the average accuracy of Wake stage classification was 83.29%, 82.11% for NREM stage and 76.73% for REM stage.