Title: A systematic review on latest approaches of automated sleep staging system using machine intelligence techniques

Abstract

Background and Objective: Sleep staging plays a vital role in sleep research because sometimes sleep recording errors may cause severe problems like misinterpretations of the changes in characteristics of the sleep stages, medication errors, and finally, errors in the diagnosis process. Because of these errors in recordings and analysis, the sleep behavior and automated sleep staging system are adopted by different researchers with different methodologies. This study identifies specific challenges with the existing studies and highlights certain points that support the improvement of automated sleep staging-based polysomnography signals. Methods: This work provides a comprehensive review of an automated sleep staging system, which was contributed by the different researchers in the recent research developments using Electroencephalogram, Electrocardiogram, Electromyogram, and combinations of these signals Results: Our review in this research area shows that single-model and multi-modal signals are used for sleep staging, and also we have observed some great points from the existing methodologies: (1) It has been noticed that 30-s length of the epoch of EEG signals may not be sufficient to extract enough information for discriminating the sleep patterns but in the other hand that a 10-s and 15-s length epoch is well suitable for sleep staging, (2) due to similar characteristics on N1 and REM sleep stages, most of the traditional classification models misclassified N1 sleep stages as REM stage, which alternatively degrades the sleep stagging performance,(3) consideration of heterogeneous form signal fusions can give the improvement results on sleep staging, and (4) applying deep-learning based models like convolutional neural network (CNN) and extended short-term memory network (LSTM), deep models, combinedly to significant PSG signals can lead to more robust automatic sleep staging results. Conclusions: The review mentioned above points simultaneously improves automated sleep staging by polysomnography signals. These points can help to focus our research work from the traditional feature extraction method to systematic improvements such as automatic feature recognition without explicit features, a proper characterization of the sleep stage’s behavior, safety, and reduced cost.