Developing a new method for obstructive sleep apnea diagnosis based on machine learning

Obstructive Sleep Apnea (OSA) is a disease caused by breathlessness in sleep. Diagnosis of the disease is performed by polysomnography (PSG) device with sleep staging and respiratory scoring steps. The system structure causes many discomfort to the patient during diagnosis. Alternative systems are needed for the PSG device, which can be a solution to the inconveniences. In this thesis study, a new approach was developed to PSG device. This approach has been proven that an alternative to PSG is to create a device that is less disturbing to the patient and as reliable as PSG. In the study, a Photoplethysmography (PPG) signal from 10 individuals was used. For diagnosis, an artificial intelligence-based diagnostic algorithm is designed using PPG signal and Heart Rate Variable (HRV) derived from PPG. For the study, 86 features were extracted, 46 of PPG and 40 of HRV. Statistically, the Mann-Whitney U test was used to determine whether the extracted features were discriminatory for sleep – wakefulness and abnormal respiratory events (apnea present - absent). In addition, features are reduced by F-score property selection methods 2 times and classified. According to the statistical results, 75 of the 86 features were significant for sleep awake (p<0,05) and 58 for respiratory scoring (p<0,05). According to the classification results, the sleep classification was successfully classified with the help of ensemble classifier with 11 features, 84,93% sensitivity, 97,40% specificity and 91,09% classification accuracy. Respiratory scoring was successfully performed with 86 features with 87,78% sensitivity, 95.46% specificity and 92.54% classification accuracy. According to the results obtained in this study, it was concluded that features of the PPG signal and the HRV derived from PPG can be used in the sleep staging and respiratory scoring process and have meaningful results. The easy acquisition of the PPG signal and the derivation of the HRV from the PPG signal opens up the possibility of performing sleep staging and respiratory scoring with a single signal. In systems that can operate in real time, easy measurement and easy handling of the signal will increase the practicality of the systems.