A Systematic Review on Deep Learning Models for Sleep Stage Classification

Abstract

Sleep is a key aspect of the body's recuperation, memory integration and consolidation processes, as well as a vital part of overall health. Early discovery of sleep related disorders can help to prevent subsequent sleep problems and measuring sleep stages can be as useful tool for sleep research. Sleep analysis is based on expert led imagery and is sensitive to inter and intra-observer variability. Supporting specialists with such a Programmable Diagnostic Tool (PDT) equipped with different machine learning algorithms are useful for diagnosis of sleep problems. But used need to be educated with the technology and it will be difficult in practice. This research paper presents a systematic review of 20 studies that used various deep learning models to classify sleep stages using polysomnography datasets. According to this paper, the majority of the research is based on convolutional neural network (CNN) over electroencephalography (EEG) datasets to classify sleep stages and obtained great performance. The accuracy of feature extraction and the classification algorithms are evaluated. This study also shows that deep learning model namely CNN is effective in terms of classification accuracy. For deep learning techniques to be completely implemented as a workable PDT for scoring sleep stages in medical applications, other Polysomnography data must be included in addition to EEG datasets. This review paper covers methodologies from the recent decade, focusing the use of deep learning models in combination with some other polysomnography data like EMG, EOG, ECG and EEG for scoring of sleep-stages.