Analysis of Sleep apnea Considering Electrocardiogram Data Using Deep learning Algorithms

Abstract

Abstract Sleep is a vital component of every human being. Adequate restful and restorative sleep reenergizes the body, enhances overall health and psychological well-being. Sleep hygiene, chaotic lifestyles, disorder breathing, stress, and anxiety contribute to poor sleep quality. Obstructive sleep apnea (OSA) sleep respiratory disorder causes temporary lapses of breathing results in gasping, choking, snoring sounds during sleep. The individual does not consciously wake up, but the brain has to start breathing again which disrupts the sleep quality. Polysomnography (PSG) sleep study is employed to diagnose sleep disorders by using either in-home or laboratory-based comprehensive tests. The untreated OSA leads to deterioration in health, performance consequences with severity including daytime sleepiness, motor vehicle accidents, workplace errors, cardiovascular morbidity, and mortality. The pre-processed, interpolated and segmented ECG signal is considered for the examination of OSA. This paper focuses on three types of deep learning classifiers-based prediction models for detection of apnea from the ECG signal. The accuracy value of Long Short Term Memory model (LSTM) is 85 percent and classifier’s ability to distinguish between normal and apnea events is 0.88.The Gated Recurrent Unit (GRU) classifier and Convolution Neural Network (CNN) model have an f1- score value of 0.80. The proposed LSTM model provides the optimal performance in comparison to other deep learning models used for classification with respect to area under the curve (AUC) and accuracy metrics.