Detection of the relationship between sleep apnea and sleep stages

Sleep is an important process for human life. It is the most basic form of rest for the human body and its quality significantly affects human life. There are many diseases that can cause problems in the heart and brain caused by poor sleep quality. The simplest of these diseases is snoring and the most serious is sleep apnea. Sleep staging plays an important role in the diagnosis of sleep disorders, especially sleep apnea, that negatively affect daily life and can cause death. Sleep staging is done by evaluating the biomedical signals recorded during sleep by a sleep specialist according to the standards set by the American Academy of Sleep Medicine (AASM). These signals (Electroencephalogram (EEG), Electromyogram (EMG), Electrocardiogram (EKG) and Electrooculogram (EOG)) are recorded in sleep laboratories with the Polysomonography (PSG) device. Except for the wake phase, sleep is divided into two main parts: Non-Rapid Eye Movements (NREM) and Rapid Eye Movements (REM). The NREM phase is also divided into three parts: NREM1, NREM2, and NREM3. Every individual experiences NREM and REM stages in a certain order during sleep. The frequency of these stages during sleep gives information about the sleep quality and health of the individual. In this thesis, it was aimed to examine the relationship between sleep apnea and sleep stages after sleep staging with EEG, which is one of the signals recorded with PSG. While staging sleep, three different feature sets were created from the EEG signal. The first feature set consists of 80 features extracted from the subbands obtained by the Discrete Wavelet Transform method. The second feature set was composed of 26 features obtained using the Empirical Mode Decomposition (EMD) method, and the third feature set was composed of 26 features extracted by the Variation Mode Decomposition (VMD) method. By combining all the features in the feature sets, the fourth feature set consisting of 132 features was obtained. Finally, the most effective 91 features were selected by applying the Relief feature selection method to the fourth feature set, and the fifth feature set was created with these selected features. For each of these five feature sets, sleep staging was performed for Artificial Neural Networks (ANN), Support Vector Machines (SVM), k Nearest Neighbor Algorithm (kNN) and Bagged Tree Algorithm (IT) classifiers. The highest test success rate of 70.02% was obtained with the features extracted by EMD method and ANN classifier. By registering the highest classifier model, 9 different patients not used in staging were used for validation, and the relationship of sleep apnea disease of these 9 patients with stages was examined.