A New Approach for Identifying Patients with Obstructive Sleep Apnea Using K-Nearest Neighbor Classification

Obstructive Sleep Apnea (OSA) is a common health issue in the United States. It is defined as the repeated blockage of the upper airway, which causes the interruption of breathing during sleep. According to the Frost and Sullivan calculation, the annual economic cost of undiagnosed sleep apnea is approximately $150 billion in the United States alone. Polysomnography (PSG) is the most comprehensive assessment method for sleep apnea and involves spending a night away from home attached to many sensors in a clinic bed. As a result, diagnosing sleep apnea is inconvenient and expensive. There has been much research in recent years to find a more convenient and inexpensive approach for sleep apnea classification. This study proposes a machine learning classification algorithm that processes short periods of electrocardiogram (ECG) signals for obstructive sleep apnea detection. The effect of sleep apnea on cardiovascular variability was measured by extracting two characteristics of the ECG signal: the power of the very-low-frequency component and the standard deviation of R-R intervals. A new sleep apnea classification algorithm was developed based on K-Nearest Neighbor (KNN) supervised learning and applied to 50 recordings from subjects with OSA and healthy subjects. The designed classification model can detect OSA patients with 90% accuracy in the testing dataset. The algorithm could be used as a platform for designing any mobile application or portable embedded system for detecting OSA.