Obstructive Sleep Apnea (OSA) is one of the most common sleep disorders, and the number of patients is increasing. Polysomnography(PSG), which is performed to diagnose OSA, has limitations. Accessibility is low because sensors must be attached and measured in the laboratory, and manpower consumption is high because clinical experts monitor an average of 7 to 8 hours of sleep time. Because of the high cost of testing and the high demand for potential OSA patients compared to laboratories, many patients go undiagnosed. Therefore, we devise an OSA diagnosis method, which is an alternative to PSG, by solving the limitations of PSG, which is constrained by location and financial burden.
In this paper, we propose a respiratory state classification method based on an electrocardiogram, which is a simple biosignal, for the development of OSA pre-test tools to replace PSG and to diagnose OSA and to prioritize test subjects. We develop a respiratory state classification model using public data provided by Physionet and data provided by actual hospitals.
The data was preprocessed so that the machine learning technique could be applied for the respiratory state classification using the statistical machine learning technique, which is the first proposed method. To solve the data imbalance, the oversampling technique smote was used to improve the performance.
The second proposed method, the respiratory state classification model using the deep learning technique, detected the R-peak, calculated the R-R interval, the amplitude of the electrocardiogram, and the EDR, and used it as a feature. After that, a deep learning-based classification model mixed with CNN and LSTM was created and applied. As a result, the method using machine learning on public data showed an accuracy of 80%, and the method using deep learning showed an accuracy of 90%. The data provided by the actual hospital showed results of 69% and 80%, respectively. The method using machine learning showed the best performance as a result of comparison with existing studies using public data.
Using the ECG-based respiratory state classification method proposed in this paper, it is possible to determine the respiratory state by measuring the ECG. Therefore, it can be mounted on a wearable device capable of measuring an electrocardiogram, so that it is possible to easily grasp the breathing state during sleep. Usually, if the accuracy is more than 80%, it can be commercialized and used as a preliminary test. Therefore, if the method proposed in this paper can be mounted on a wearable device and the breathing state can be detected with the wearable device, it can be expected as an accessible OSA pre-test method.