Deep learning-based sleep apnea detection using single-lead ECG signals from the PhysioNet apnea-ECG database

Abstract

Sleep apnea, a significant medical concern affecting many individuals, is the focal point of this research, which investigates the potential of machine learning and deep learning techniques, particularly in detecting Obstructive Sleep Apnea (OSA). The study's results highlight the effectiveness of models like 1D-CNN in processing electrocardiogram (ECG) signals without requiring manual feature extraction. Notably, our 1D-CNN model achieves an impressive accuracy of 88.36% using 6,000 features, demonstrating strong recall and F1 score performance. Conversely, our Random Forest (RF) model attains a commendable accuracy of 82.23% with just seven features, showcasing high precision and F1 score. However, the K-Nearest Neighbors (KNN) model, characterized by high precision, displays lower recall and specificity, indicating a propensity to classify all data as positive cases. These findings underscore the potential of machine learning and deep learning techniques to enhance sleep apnea detection, offering valuable insights for the diagnosis and management of this critical medical condition.