A Research Of Detecting Sleep Apnea Based On Electrocardiogram

Sleep apnea has gradually attracted more and more attention. The long-time of sleep disordered breathing events will not only cause the body disease, but also lead to the impact on life and work. PSG has lots of disadvantages such as the multi-Guide, expensive, uncomfortable and so on. With the progress of human society, in order to make more and more people detect their sleep state, the portable sleep monitoring system is on the top of the list. The acquisition of the single lead ECG signal is simple to realize, which could just use two electrodes to obtain ECG signal. According to previous work, ECG signal contains a wealth of physiological information to study the chara cteristics of sleep apnea.In this paper, the purpose is to find out whether the ECG features could be used as the indicators of sleep apnea based on the previous research of ECG signal. Using these parameters which are calculated from ECG features builds detecting algorithm. First, the ECG data should be divided into 15 s epoch. Then is the ECG data processing, including using notch filter to remove the power frequency interference and using low pass filter to reject the high frequency noise. Feature parameters are extracted from the each epoch and the average value of each feature should be calculated as the sign of detecting sleep disorder in short time. The features are extracted by ECG detecting algorithm including the detection of QRS and T wave. The breathing extraction algorithm is built by using R feature. Then the parameters of these signs are extracted like RR interval, RR amplitude, TT interval, TT amplitude, the angle of QRS and HRV. And obtain their means and variances as the classified features. A classification model is built based on support vector machine which adopts RBF. Two processes are included. One is the choice of SVM kernel, the other is the optimization of kernel function parameters. The test data from Apnea database is used to test the performance of classified algorithm and test the performance of using different signs. The result shows that the mean value of RR interval, the mean value of TT interval and the mean value of inter quartile all have the better accuracy, sensitivity and specificity than other features. Using the combination of these three signs evaluates the sleep apnea which has a practical value in wearable devices.This paper chooses the most relevant characteristics by studying the time domain of ECG signal. Then using these features as the classified features detects sleep apnea. These theoretical research es are the foundation to realize undisturbed detection of sleep apnea.