AI-based Prediction Model for Incident of Obstructive Sleep Apnea Using ECG Signals: Utilization of MrOS

Background: In recent years, obstructive sleep apnea (OSA) has been diagnosed using various devices and algorithms, based on diverse physiological signals.However, few studies have assessed prediction models and methods for identifying the incidence of OSA.We propose a novel method using an artificial intelligence (AI)-based prediction model for the incidence of OSA using electrocardiogram (ECG) signal.Methods: The proposed model was implemented using a deep convolutional neural network (CNN) to predict the incidence of OSA using a single-lead ECG recording from polysomnography (PSG).The AI-based prediction model was designed and constructed as a 5-layer CNN structure consisting of convolutional, pooling, and fully connected layers.The proposed AI-based prediction model was assessed using an open-access PSG database from Osteoporotic Fractures in Men sleep study.Results: All datasets consisted of single-lead ECG signals from 55 participants (30 controls, 25 patients with OSA) aged 65 years.The evaluation performance of the AI-based prediction model yielded accuracy rates of 85.3%, 82.2%, and 85.7%, respectively, for the training, test, and validation sets. Conclusion:The results demonstrate the possibility of AI-based prediction of OSA incidence from a single-lead ECG for sleep monitoring in elderly people, and can help in the early and straightforward prevention of OSA.