Diagnosis of sleep apnea using artificial intelligence methods from ECG signals

Sleep is a repetitive state in which living creatures react minimally or unresponsively to external stimuli. Diseases such as chronic insomnia, eating disorders, restless legs syndrome and sleep apnea are among the diseases that affect sleep quality and cause serious problems. The syndrome of recurrent episodes of cessation of breathing for 10 seconds or longer because of narrowing or obstruction of the upper airways called sleep apnea, and the most common is Obstructive Sleep Apnea (OSA). In this thesis, sleep apnea detected with using machine learning and deep learning algorithms from ECG signals. In the first stage, the ECG signal divided into segments and R peaks determined and each segment classified that segments sleep apnea or not. In the second stage, each of the 35 ECG recordings that were classified based on segments classified that records OSA or not. Seven different classifiers used as machine learning algorithms and three different classifiers used as deep learning algorithms. In classification based on segments, the Convolutional Neural Network (CNN) algorithm that obtained the highest results with 89.11% accuracy; 94.31% sensitivity; 80.72% selectivity; 0.89 F1 score and 0.87 AUC values. In the results of OSA and normal classification of each recording according to the Apnea-Hypopnea Index, the CNN model was the algorithm that obtained the highest results with 97.14% accuracy; 100% selectivity; 95.65% sensitivity; 0.97 F1 score and 0.98 AUC value.