Detection Of Snoring And Sleep Apnea Syndrome Based On Hybrid Neural Network

Sleep apnea is a serious disease that causes continuous interruption of sleep and transient hypoxemia and elevated carbon dioxide due to repeated periods of respiratory arrest,and obstructive sleep apnea(OS A)is one of the most common types.A common type is caused by a recurring throat or upper airway closure during sleep.Polysomnography(PSG)is the "gold standard" for OSA diagnosis,however,it is difficult and expensive to perform PSG diagnosis for patients.Therefore,there is an urgent need for a rapid and effective method for detecting obstructive sleep apnea events.The snoring is the basic signal of obstructive sleep apnea,carrying information about upper airway obstruction,accurately analyzing and assessing the snoring of potential OSA patients,and has a good guiding role in the diagnosis of OSA.Therefore,this paper starts with the acoustic analysis of the snore,and proposes the detection algorithm of OSA.The main work of this paper is as follows:First,only the Mel Frequency Cepstral Coefficents(MFCC)feature is used,and the model is built based on the Stacked Autoencoder(SAE).By classifying different events from the sleep sound record(Snore,OSA and silence)to evaluate the performance of the proposed model.Then,the feature and model structure are deeply explored,and the linear predict coefficient(LPC)feature is added and the model is built based on the hybrid neural network(HNN).Comparing the classification results of the model with the correct annotations,the proposed model can achieve 90.65%sensitivity for the detection of snoring events,90.99%sensitivity for OSA detection,and 90.30%sensitivity for silent events.The results show that the model can provide a basis for the diagnosis of OSA.