An ANN-Based Detection of Obstructive Sleep Apnea from Simultaneous ECG and SpO₂ Recordings

Abstract

Obstructive sleep apnea (OSA) is one of the most common sleep disorders characterized by a disruption of breathing during sleep. This disease, though common, goes undiagnosed in most cases because of the inconvenience, cost, and/or unavailability of opting for polysomnography (PSG) and a sleep analyst. Many researchers are working on devising an unsupervised, cost-effective, and convenient OSA detection methods which will aid the timely diagnosis of this sleep disorder. Commonly used signals to detect OSA are ECG, EEG, pulse oximetry (SpO₂), blood oxygen saturation (SaO₂), and heart rate variability (HRV). In this work, an attempt to detect the OSA using simultaneously acquired ECG and SpO₂ signals has been presented. Various features from the RR intervals of ECG, and a couple of features—namely, CT90 and delta index—from the SpO₂, were extracted as indicators of OSA. The features were then fed to a trained artificial neural network (ANN) which classified the signals as OSA positive or OSA negative. The proposed technique boasts a very high accuracy of 98.3%, which is superior to other competing techniques reported so far.