An Automated Scoring for REM and N3 Stage Using Wavelet Filters and Support Vector Machine with PSG Signal

Sleep architecture and sleep quality have constantly been engaging topics for sleep researchers. One of the fundamental foundations for any sleep study is to build the hypnogram based on Rechtschaffen and Kales (R&K) or American Academy of Sleep Medicine (AASM) rules. This study focuses on establishing a completely automated scoring sleep stages system. The primary purpose is to classify two important sleep stages that have a decisive role in improving sleep quality: Rapid Eye Movement (REM) and the deep sleep stage (N3). The primary method used in this study is applying wavelet filters combined with fast Fourier transform (FFT) and Infinite impulse response (IIR) filters to extract important features such as the ratio of Delta and slow-wave activities, characteristic of rapid eye movements, mean average power and support vector machine (SVM) for training and classifying to three classes. The first data set was used to extract polysomnography (PSG) signals, including electroencephalography (EEG), electrooculography (EOG), and electromyography (EMG) collected from the edfx database from physionet open source. The second data set used in this study is 10 experimental measurement data at Biomedical Engineering Labs, Faculty of Applied Science, Ho Chi Minh City University of Technology (HCMUT). The results of this study show that data classified into three classes achieved an accuracy of 93%. The new contribution of this research is the establishment of a simple and automated process for polysomnography signals based on the characteristics of each sleep stage. The focus is on the classification of two states, REM and N3, which play an important role in assessing adult sleep quality. The results from this study can be used to develop studies that improve sleep quality, such as increasing the deep sleep phase and generating important stimuli in the REM stage.