Recognition of Obstructive Sleep Apnea from a Single Lead Electrocardiogram Signal Using a ResNet Model

Obstructive sleep apnea (OSA) is a prevalent sleep disorder that induces respiratory disturbances during the state of sleep. Furthermore, it is associated with several comorbidities, such as cardiovascular ailments like coronary heart disease and hypertension. Nocturnal polysomnography (PSG) is a clinical diagnostic modality utilized for the diagnosis of OSA. This method is associated with significant costs and discomfort for patients, as it necessitates manual interpretation by trained professionals and is time-consuming. Methods utilizing electrocardiogram (ECG) signals for the diagnosis of OSA have been proposed as a means of mitigating associated challenges. Nevertheless, the vast majority of the currently available solutions are based on feature engineering, which requires a large amount of specialized knowledge and skill. The present study introduces an innovative technique for the categorization of OSA through the utilization of a ResNet model on a solitary-lead ECG signal. In order to train and test the proposed model, the PhysioNet Apnea-ECG database was used throughout the process. ReseNet18 and ResNet50 are both deep learning models, and they are used to predict how well the model will perform. For a batch size of 256, after 2, and before 2, the ResNet 18 proved its maximum performance by achieving an accuracy of 89% and a recall of 89%. In addition, the ResNet 50 was able to attain its maximum performance by doing the following: obtaining an accuracy of 89%, a precision of 89%, an F1-score of 89%, and a recall of 89% for the batch size of 512, after 1, and before 1.