Study On Classification And Prediction Of Sleep Apnea Hypopnea Syndrome Event Based On Respiratory Signal

Sleep apnea hypopnea syndrome(SAHS)is a kind of disease that occurs during sleep.It affects people’s sleep safety and can induce various diseases such as hypertension,coronary heart disease and arrhythmia.Polysomnography is the gold standard for detection of SAHS.But it is complicated,time-consuming and expensive in its practical application,which makes it impossible to be applied in a wide range.Therefore,it is necessary to propose a simple and effective method to detect and predict SAHS events.The existing algorithms are studied for detection and prediction of SAHS events.Then,an algorithm is proposed using traditional classification methods and deep learning classification methods for detection and prediction of SAHS events based on respiratory signals.The MIT-BIH Apnea Datebase is used to verify the algorithm including 8396 samples of 8 patients in minutes.The main work of this paper is as follows:(1)Characteristics of respiratory signals in SAHS events are studied and signals are then classified based on feature vectors using traditional classification methods.After data preprocessing,the time-domain characteristics such as variance and the number of clinical zero crossings,the frequency-domain characteristics such as energy and maximum value of wavelet coefficients,the nonlinear characteristics such as fractal dimension and sample entropy are extracted to form a 6-dimensional feature vector.Support vector machine and BP neural network are used for SAHS classification.The results show that the average accuracy of the two methods is 71.2% and 84.5%.According to the characteristics of classification results,the method of morphological filtering is introduced to improve classification results of BP neural network.It improves the classification results by 85% and the accuracy rate is increased to 86.3%.(2)Classification of respiratory signals uses convolutional neural network method in deep learning.Based on the classical convolutional neural network Alex Net,a network suitable for respiratory signals is constructed according to clinic requirements.The respiratory signals are input into the convolutional neural network after down-sampling,normalization and segmentation.The average recognition rate of SAHS events reaches 92.12% by this method,which shows better classification performance than the other two traditional classification methods.(3)Prediction algorithm of respiratory signals is proposed using Multilayer perceptron(MLP)neural network based on Ada Boost technology.Firstly,respiratory signals from normal and SAHS patients are predicted using linear prediction,Kalman filtering and BP neural network.Then the prediction algorithm using MLP neural network based on Ada Boost technology is proposed.The evaluation parameters of root-mean-square-error(RMSE)and cross-correlation coefficient are calculated respectively.The results show that the proposed prediction algorithm has the best prediction effect because the RMSE of it is lower than 0.03 and the cross-correlation coefficient is higher than 0.9.A theoretical basis for the classification and prediction of SAHS events is provided by the results of this research.Also,the given prediction algorithm in this paper lays a foundation for real-time warning and active intervention of SAHS in sleep respiratory monitoring in the near future.