Feasibility analysis of sleep apnea detection from heart and respiratory sounds

Sleep apnea syndrome is a serious disorder which cause disrupt of the sleep pattern and it can be defined cessation of mouth and nasal airflow for at least 10 seconds during the sleep. If it is untreated, it can be cause serious health problems like heart attack and stroke. Polisomnography (PSG) considered the "gold standard" for diagnosis of sleep apnea syndrome and other sleep disorders. But PSG has some disadvantages. It is expensive, time consuming and needs qualified technician and equipment. These disadvantages of PSG leads to find new techniques for detection of sleep apnea. For the first time in literature, in this study, it has been investigated whether detection of sleep apnea is possible or not with features extracted from heart and respiratory sounds with the motivation of having obtained success on detection of sleep apnea with features extracted from only ECG and having reported that the heart sounds exhibited strong morphological variability during respiration. For this purpose, 20 objects' heart and respiratory sounds were recorded simultaneously with PSG. Signal regions with apnea or not have been apnea identified by PSG device. And then these signal regions have been taken as a reference for the classification of sleep apnea. For the classification, K nearest neighbor algorithm and support vector machines were used in this work. The classifications were done with the combination of the time and frequency domains parameteres obtained from heart and respiratory sounds, and the best classification results were obtained when using the time and frequency domains parameteres of heart and respiratory sounds together with using the K nearest neighbor algorithm. But even in the best case, 100 % specifity and 48 % sensitivity were obtained. As a result, it was decided that, the snoring, which is the most common sign of sleep apnea patients, makes difficult to classify and the features obtained from the heart and respiratory sounds were not sufficient for detection of sleep apnea successfully.