Sleep Stage Identification Based on EEG Signals Using Parallel Convolutional Neural Network and Recurrent Neural Network

Abstract

Sleep quality is essential to health, informed by the sleep stage. In other words, identifying the sleep stage can detect the possibility of sleep disorders. The standard carried out in medicine is Polysomnography (PSG) which consists of many devices. A simple one-channel Electroencephalogram (EEG) signal is one device that can identify sleep levels in humans. It means minimizing additional sleep disturbances. EEG captures electrical activity in the brain using electrodes. Identifying sleep levels is challenging as it usually uses a pair of channels. Many studies have discussed sleep disorders using several well-known methods such as Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN). This paper proposed parallel CNN-RNN methods that provide advantages in identifying EEG signals due to the characteristics of CNN, which processes features on the channel, and RNN, which processes sequence data. The parallel CNN-RNN method identified five sleep stages: Wake, non-REM 1 (N1), non-REM 2 (N2), non-REM 3 (N3), and Rapid Eye Movement (REM). Dataset recorded from the Sleep-EDF dataset with several EEG signal channels. The Wavelet feature was used to extract the features contained in the signal. The experimental results of the two EEG channels produced high accuracy values, which are 90.13 % for the Fpz-Cz channel. This proposed model using parallel CNN-RNN achieved higher performance based on single-channel EEG.)