Design Of An Assisted Diagnosis System For Sleep Apnea Syndrome Based On ECG And SpO2

Sleep Apnea Syndrome(SAS) is a common sleep disorder that poses a serious threat to human health.Prolonged occurrence of sleep apnea can induce cardiovascular diseases such as hypertension and heart failure.Currently,polysomnography is a commonly used tool for detecting SAS.However,polysomnography(PSG)as the gold standard for the diagnosis of SAS is costly,time-consuming and uncomfortable.Therefore,research on the automatic diagnosis of SAS using a small number of channels of physiological signals is important to facilitate the development of SAS detection techniques.In this paper,the automated SAS diagnostic algorithm was developed based on traditional machine learning techniques and deep learning techniques respectively,using a combination of electrocardiogram(ECG)and saturation of peripheral oxygen(SpO2)signals to improve the accuracy of diagnosing sleep apnea.The main research contents include:(1)In this study,a feature fusion-based algorithm for automatic SAS detection was proposed.A multimodal approach was used to fuse ECG and SpO2 signals at the feature level,and a recursive feature elimination method combined with random forest was used for feature selection.Finally,sleep apnea was detected using Support Vector Machine,Logistic Regression,K-Nearest Neighbor,and Random Forest classifiers,respectively.The experimental validation was conducted on the Apnea-ECG database.The experimental results showed that the best results were obtained by using the RF classifier in distinguishing apnea from normal segments,achieving 97.5% accuracy,95.9 sensitivity,98.4 specificity and an AUC of 0.992.In addition,the combination of ECG and SpO2 signals improves diagnostic performance compared to using only a single signal to detect sleep apnea.(2)In this study,the interaction between ECG and SpO2 signals in diagnosing SAS was analyzed.To this end,the feature selection algorithm was used to reveal the internal links between the features and the KW-ANOVA test was used to analysis the features for significant differences.The results of feature selection show that ECG signal and SpO2 signal are complementary in the task of diagnosing SAS.The results of the KW-ANOV A test shows that for all the 13 selected features,p << 0.01,which means that the selected features are statistically significantly different in discriminating between normal and SA classes.(3)In order to simplify the process of diagnosis and achieve end-to-end diagnosis of SAS,a model combining a 1-dimensional convolutional neural network(CNN)with a long short term memory network(LSTM)was proposed.The proposed model uses the RR interval signal extracted from the ECG signal and the SpO2 signal together as input for the detection of sleep apnea.The experimental validation was conducted on the Apnea-ECG database.The experimental results show that the model achieves 98.3% accuracy,96.1%sensitivity and 99.1% specificity and performs slightly better than the feature fusion-based algorithm.(4)In this study,a web-based SAS-aided diagnosis system was designed.The system communicates in real time between the web side and the server side.Then,a 1-dimensional CNN-LSTM model was called on the server side to detect SAS.Finally,the results were returned to the web page for display.In summary,the SAS automatic diagnosis algorithm proposed in this paper has excellent performance,and the system designed can play an auxiliary screening role in the detection of SAS.This study provides a reference for the development of intelligence in the field of sleep monitoring.