Obstructive Sleep Apnea Detection from ECG Signals with Deep Learning

Among the newer problems that have been linked to heart disease, hypertension, sleeplessness, and other ailments in recent years is sleep related disorders. One such newly arising problem is sleep apnea. The three predominant types are obstructive, central, and mixed. Obstructive sleep apnea (OSA) is becoming highly prevalent. It affects individuals of all ages, with the senior citizens being the majority affected. OSA seems to have a significant influence on the healthy sleep cycle, which contributes to a variety of cardiovascular diseases. Polysomnography (PSG) is a customary or traditional approach to identify sleep related disorders and numerous alternatives have already been developed in recent years to substitute the classic method due to its underlying complexity. However, the precision of these tools is typically inadequate for clinical evaluation. Numerous researchers have explored the utilization of deep learning algorithms to detect obstructive sleep apnea (OSA). Although these investigations have yielded encouraging outcomes, they do have certain constraints. The absence of standardized protocols and guidelines for acquiring data, preprocessing it, and extracting features in OSA detection using ECG signals is evident. Additionally, only a limited number of studies have been conducted on relatively small datasets, potentially lacking comprehensive representation of the diverse OSA-affected population. Hence, this research study introduces novel methods for classifying sleep apnea recordings using three types of classifiers, namely ANN, CNN, and LSTM. These classifiers were trained and evaluated using Python programming language on sleep data collected from individuals both with or without the presence of OSA. Findings of our automatic detection method indicate that it is capable to identify stages of sleep disorders having 87% or more precision. Additionally, this suggested system has the potential to serve as a foundation for creating a tool for screening OSA in the future.