Obstructive Sleep Apnea Detection Using Sleep Architecture

Abstract

Obstructive sleep apnea (OSA) is a common disease characterized by repeated episodes of upper airway obstruction that results in cessation of airflow during sleep. Early diagnosis of OSA is essential so that early intervention can reduce the risk of cardiovascular disease, metabolic disorders and neurocognitive dysfunction. Sleep architecture is related to OSA. In this paper, the patient's sleep stages and their transitions relationship are used as features to propose a machine learning-based OSA detection method. The key parameters are screened through statistical analysis. Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM) are used to establish classification models. The whole of results show that XGBoost has a better performance with the area under curve of 0.9128, and find that age, the percentage of N1 sleep stage, the percentage of N3 sleep stage, one-step transition pattern of N2→N1 and total number of transitions play important roles in identifying OSA patients from normal subjects.