Study On The Method Of Automatic Detection Of ECG-SAS Based On RBF Neural Network

Sleep Apnea Syndrome (SAS) is often called "dream killer". It is a common disease thathas serious threat to human health in nowadays. Polysomnographic (PSG) monitoring is thegold standard for clinical diagnostic. However this method has many shortcomings such asthe complexity of the implementation process, expensive and poor precision, etc. whichcause its popularization and application are difficult, and many patients can’t receivediagnosis and treatment in timely. It is significant to study of a non-invasive, high-precisionautomatic detection method which can replace PSG. Among all the methods at present, ECGindirect detection method is the most representative method, which is easier to implement,and has higher detection accuracy.This paper presents and designs an automated method for detecting ECG-SAS based onartificial neural network. Eliminating ECG baseline drift with wavelet decomposition andreconstruction method, and removing frequency interference and ECG interference bythreshold value method. By accurately locating the R-wave based on Modulus maxima,fulfill the RR intervals correction, which reduces the problems of R-wave undetected anderror detection effectively, and enhances the concentration ratio of RR intervals discretepoint. Extracting a series of time-domain and frequency-domain characteristic parametersthat can characterize the features of SAS based on the study of RR intervals and heart ratevariability. And establishing ECG-SAS automatic detection model via BP and RBF neuralnetworks. By accuracy detection evaluates the sensitivity and the specificity. Finally findingthe optimal model, which realizes the accurate and automatic detection of SAS..The BP neural network model is used to test the training and test sets of Apnea-ECGdatabase provided by Physionet. The accuracy of the training sets respectively is82.85%,The accuracy of the test sets is80.00%. Then using the RBF Neural Network tests test thetraining and test sets. The accuracy of the training sets respectively is94.28%, The accuracyof the test sets is85.71%.The results show that the Neural Network method based on ECG-RBF proposed in thispaper has the advantages of simple operation and high accuracy and efficiency.