Sleep Stage Classification Based on EEG, EOG, and CNN-GRU Deep Learning Model

Abstract

This paper presents a CNN-GRU deep learning model for classifying sleep stages. The Conventional sleep stage scoring method is a visual classification process, based on a set of biomedical signals such as Electroencephalogram (EEG) and Electrooculogram (EOG), where high human intervention is required. In this study, we proposed a deep neural network involving convolutional neural networks and gated recurrent units, to automatically extract the most appropriate features and sequence trends of PSG signals, without utilizing hand crafted features for scoring sleep stages. The proposed model, which uses multiple PSG channels, was evaluated using two data sets collected from 184 patients and 70 healthy subjects. The proposed multi-channel model showed 91.9 % of overall accuracy, while recall, precision, and f1 measures were approximately 92 % for patients. For healthy subjects, the multi-channel model showed 89.3 % overall classification accuracy. Recall, precision, and f1 measures showed approximately 89 %. The main model was adapted to utilize with a single EEG channel configuration, which yields 4 single-channel models for each data set. Therefore, the proposed model is capable of performing sleep stage classification using a single EEG channel without altering the model architecture.