Performance analysis of sleep apnea detection and categorization using deep learning and optimization techniques

Abstract

One of the most important aspects of human health is sleep, and conditions such as sleep apnea can severely disturb the quality of sleep, which can lead to a variety of health concerns. Polysomnography (PSG) is the sleep study test commonly used to diagnose breathing related sleep disorder by collecting various bio signals from the human body. On the other hand, PSG requires a significant number of resources, and it is time consuming, expensive, and can cause discomfort for patients since sensors are wrapped around the patients to collect the data, which demands the solutions for more automatic and effective therapy methods. This study addresses the need for enhanced diagnostic tools for sleep apnea. The multi-phase strategy approach is employed for the detection, for assessing severity, and for event classification of sleep apnea by utilising deep learning framework along with the optimization techniques. newlineThe utilization of electrocardiogram (ECG) signals for the identification of sleep apnea is the primary interest and ECG data are pre-processed, segmented, and interpolated. They are investigated by three models, namely Convolutional Neural Networks (CNN), Gated Recurrent Units (GRU), and Long Short-Term Memory (LSTM) networks. It was proved that the LSTM model had higher performance, with an accuracy of 85% and an Area Under the Curve (AUC) of 0.88. This demonstrates that the model is able to accurately differentiate between apnea episodes and non-apnea events. Even though they were effective, the CNN and GRU models produced F1-scores that were lower than 0.80. newline