An interactive, real-time, high precision and portable monitoring system of obstructive sleep apnea

摘要

Obstructive sleep apnea (OSA) is the most common type of sleep apnea which is defined as the suspension of breathing. OSA is generally caused by complete or partial obstruction of airway during sleep, making the breathing pattern irregular and abnormal for prolonged periods of time. Apnea can contribute to a variety of life threatening medical conditions, and can be deadly if left untreated. Nowadays, out of 18 to 50 million people in the US, most cases remain undiagnosed due to the cost, cumbersome and resource limitations of overnight polysomnography (PSG) at sleep labs. Currently PSG relies on a doctor's experience. In order to improve the medical service efficiency, reduce diagnosis time and ensure a more accurate diagnosis, a quantitative and objective method is needed. In this dissertation, an innovative method in characterizing bio-signals for detecting epochs of sleep apnea with high accuracy is presented. Three data channels that are related to breath defect; respiratory sound, ECG and SpO2 are investigated, in order to extract physiological indicators that characterize sleep apnea. An automated method was used to analyze the respiratory sound to find pauses in breathing. Furthermore, the automated method analyzed ECG to find irregular heartbeats and SpO 2 to find rises and drops. The system consists of three main parts which are signal segmentation, features extraction and features classification. Feature extractions process is based on statistical measures. Features classification process is learned through Support Vector Machines (SVMs) and Neural Network (NN) classifiers. Moreover, a preprocessing technique is carried out to distinguish the R-wave from the other waves of the ECG signal. The approach presented in this dissertation was tested using downloaded polysomnographic ECG and SpO2 data from the Physionet database. In addition, to identifying sleep apnea using the acoustic signal of respiration; the characterization of breathing sound was carried by Voice Activity Detection (VAD) algorithm. VAD was used to measure the energy of the acoustic respiratory signal during breath and silence segments. From the experimental results for the three signals, it was concluded that the precision of classifying sleep apnea has an accuracy of 97%. This result offers a clinical reference value for identifying OSA instead of expensive PSG visual scoring method which is commonly used to asses sleep apnea, and could reduce diagnostic time and improve medical service efficiency.