Deep Learning for Single-Channel EEG Signals Sleep Stage Scoring Based on Frequency Domain Representation

Abstract

Sleep is vital to the health of the human being. Accurate sleep stage scoring is an important prerequisite for diagnosing sleep health problems. The sleep electroencephalogram (EEG) waveform shows diverse variations under the physical conditions of subjects. To help neurologists better analyze sleep data in a fairly short time, we decide to develop a novel method to extract features from EEG signals. Traditional sleep stage scoring methods typically extract the one-dimensional (1D) features of single-channel EEG signals. This paper is the very first time to represent the single-channel EEG signals as two-dimensional (2D) frequency domain representation. Comparing with similar currently existing methods, a deep learning model trained by frequency domain representation can extract frequency morphological features over EEG signal patterns. We conduct experiments on the real EEG signals dataset, which is obtained from PhysioBank Community. The experiment results show that our method significantly improved the performance of the classifier.