Hybrid RNN-based classification of Obstructive Sleep Apnea using single-lead ECG Signals

Abstract

Respiratory sleep disorders affect millions of people, with Obstructive Sleep Apnea being one of the most prevalent. Obstructive Sleep Apnea sufferers are often unaware of their illness, causing cardiovascular and neurological problems. Relaxation of the muscles that support the tongue and soft palate causes Obstructive Sleep Apnea. When these muscles relax, the patient’s airway constricts or closes, resulting in a brief cessation of breathing. Polysomnography is one of the tests used to diagnose Obstructive Sleep Apnea. While the patient is sleeping, they will be attached to technology that will monitor their heart, lungs, and brain activity, as well as their breathing patterns, leg movement, arm movement, and blood oxygen levels. Despite attempts to breathe, polysomnography reveals repeated instances of breathing delays. The majority of patients are untreated due to the difficulties caused in performing polysomnography. Using algorithms for machine learning, a number of researchers devised a variety of solutions to this issue. In the proposed work, detection of Obstructive Sleep Apnea was done by the integration of Long-Short Term Memory (LSTM) and the Gated Recurrent Unit (GRU) method. In order to validate the model, the suggested procedures made use of real-life clinical examples taken from the PhysioNet Apnea-ECG database, using thirty-five overnight sessions for Hybrid RNN (LSTM + GRU) attains 89.5% accuracy, 89.6% sensitivity, and 90.2 % percent specificity, demonstrating the efficacy of the presented method.