Utilizing Machine Learning Algorithms of Electrocardiogram Signals to Detect Sleep/Awake Stages of Patients with Obstructive Sleep Apnea
Abstract
Obstructive Sleep Apnea (OSA) is a respiratory-related disease that occurs during sleep. The diagnosis of OSA is made by a specialist doctor according to the records obtained with the polysomnography device. However, the diagnostic process is quite troublesome. More than 30 signals are recorded for diagnosis. This may cause discomfort to the patient during the night. For the diagnosis, sleep staging and respiratory scoring are performed with the records collected via the polysomnography device. With sleep staging, the patient's sleep and awake times are determined and respiratory scoring is used to detect abnormal respiratory events that occur during sleep. If more than 5 abnormal respiratory events occur per hour, the individual is diagnosed with OSA. Due to the fact that this process is laborious and uncomfortable to the individual, practical diagnostic methods are needed. In this study, an easy and practical measurement system for sleep staging, which is an important step in the diagnosis of OSA, will be proposed. According to this system, sleep/awake state is determined by electrocardiogram signal (ECG) and a machine based method. ECG records obtained from two individuals will be used in the study. First, the ECG signal will be cleared from the noise by digital filters. It will then be divided into 30 s epochs for sleep staging. From each separated epoch, 25 features will be extracted and features associated with sleep/awake will be selected with the help of feature selection algorithms. Determined features will be classified by Support Vector Machine which is a machine learning method and the system performance will be tested. In the preliminary studies, it was determined that 25 properties were related to sleep/awake stages and that the classification performance was approximately 85%. In light of this information, it is thought that a system based on machine learning can be developed for the detection of sleep/awake stages using ECG signals for the diagnosis of OSA.