Deep Learning Model for Sleep Disorder Detection

Abstract

Sleep disorders are a health issue of significant concern in the world influencing millions of citizens around the world. Insomnia, obstructive sleep apnea (OSA), narcolepsy, and restless legs syndrome are disorders that may trigger serious health issues, such as cardiovascular condition, metabolic disorders, and cognitive loss. Conventional diagnostic procedures are based on polysomnography (PSG), the practice of the measurements of several physiological parameters at the time of sleeping. Though PSG provides consistent findings, it requires expensive equipment, special clinical conditions and manual processing on the results by sleep specialists thus making it not feasible to conduct massive screening. Recent developments in artificial intelligence and deep learning have made the automatic processing of physiological signals simple to achieve medical diagnosis. In this research article, a deep learning model is suggested to be used in automated recognition of sleep disorders using physiological data, including electroencephalogram (EEG), electrocardiogram (ECG), and respiratory data. The given model includes Convolutional Neural Networks (CNN) as an object of spatial feature detection and Long Short-Term Memory (LSTM) networks as objects of detection of the time dependencies of the sleep patterns. On standard sleep data, training and evaluation are performed. Empirical experiments have shown that CNN-LSTM structure significantly enhances the accuracy of classification in comparison to the classical machine learning models, such as Support Vector Machines and the Random Forests. High accuracy and performance of the system in the detection of abnormal sleep patterns are achieved. Besides, the given solution will help clinicians to diagnose early, reduce the number of manual tasks, and monitor sleep in real-time provided people wear wearable healthcare devices.