Classification of pulse transit times in healthy individuals with osas patients using the machine learning method

Obstructive Sleep Apnea (OSAS) is the most common type of sleep apnea. It is a disease that can only be diagnosed in advanced stages because it is confused with snoring while it is not examined medically. For the diagnosis of sleep apnea, polysomnographic signaling (PSG) is used in sleep laboratories. PSG signal data include many polysomnographic signal data such as EEG (Electroencephalography signal data), Respiratory Sound, Carbon Dioxide Measurement, EMG (Jaw surface Electromyography signal data), PTT (Pulse Transit Time). In this study, by using the PTT parameter obtained at the stage of recording of the PSG signals, the classification process was made by deep learning methods which are among the methods of learning machines and healthy neural networks and Naive Bayes methods. In the application, Convolutional Neural Networks (ESA) were used as the network structure. The fc6 and fc7 layers of the AlexNet and VGG-16 structures from pre-trained network structures were compared. Support vector machines (SVM) and k-nearest neighbor algorithm (kNN) were used for classification. The accuracy, sensitivity and specificity values were calculated by using the cross-validation process in MATLAB 2017b program. Accuracy rates for SVM were AlexNet fc6 92.64%, AlexNet fc7 92.60%, VGG-16 fc6 92.78%, VGG-16 fc7 92%; For kNN, AlexNet fc6 was found 90.72%, AlexNet fc7 90.81%, VGG-16 fc6 91.72%, VGG-16 fc7 92.18%. At the same time, the accuracy value of artificial neural networks is 90% and the accuracy rate of Naive Bayes is determined as 89.78%.