Research On Prediction Model Of Sudden Cardiac Death Based On ECG Signal

Cardiovascular disease has always been the leading cause of death worldwide over the years.Sudden cardiac death is the main cause of death from cardiovascular disease.In our country,more than 500,000 people die from sudden cardiac death every year.In addition to the risk of sudden cardiac death in people with cardiovascular diseases such as coronary heart disease and cardiomyopathy,a growing number of people without cardiovascular diseases suffer from sudden cardiac arrest due to sudden occurrence of malignant ventricular arrhythmia,which will lead to sudden cardiac death if they are not rescued in a few minutes.Due to the characteristics of complex causes,strong suddenness and high mortality of sudden cardiac death,early and accurate prediction of sudden cardiac death can gain more preparation time and rescue time for medical staff,thereby improving the survival rate of patients.Based on the non-invasive and easy-to-obtain electrocardiography(ECG)signals,this paper extracts the relevant ECG features and constructs a prediction model of sudden cardiac death.The research content of this paper mainly includes the following parts:(1)Detecting the QRS complex and T wave of the ECG signal.After cutting the ECG signal and filtering it based on discrete wavelet transform,the QRS complex are detected through continuously updated adaptive threshold and multiple decision rules based on waveform slope and interval.On the premise of accurate identification of R-wave peak,the position of T-wave end and T-wave peak can be determined by sliding a fixed window in the retrieval interval based on RR interval and comparing the integral area of waveform.(2)Constructing a set of ECG features for predicting sudden cardiac death.After the ECG signal waveform is detected,a variety of morphological features based on the waveform interval and amplitude are extracted as the initial features for predicting sudden cardiac death.The mean,standard deviation and approximate entropy of these initial features are separately counted to reduce the dimension of the features.Then the feature scaling process is performed to form the final ECG feature set,which is used to train and test the prediction model.(3)A prediction model of sudden cardiac death based on support vector machine is proposed in this paper.The ECG feature set of sudden cardiac death is divided into four categories according to the four time periods: 20～30 min,30～40 min,40～50 min and 60～70min before the occurrence of sudden cardiac death.Sudden cardiac death ECG features in the four prediction periods and the same normal sinus rhythm ECG features constitute four data sets.Four prediction models based on support vector machine were constructed with four data sets to compare the prediction accuracy of ECG signals in different prediction periods for predicting sudden cardiac death.An improved grid search method is used to find the optimal parameters of the models.The experimental results show that the ECG feature set selected in this paper can advance the prediction time to less than one hour before the occurrence of sudden cardiac death.The predictive sensitivity of ECG signals from 20 min to 70 min before the occurrence of sudden cardiac death is over 86%,and the accuracy is over 90%.The average accuracy of the four models with different prediction periods are 94.84%,92.10%,90.65%and 90.08%,respectively.The closer to sudden cardiac death,the higher the accuracy of prediction.Compared with the existing prediction models,the proposed model can achieve an earlier prediction of sudden cardiac death with a higher accuracy,and the generalization ability of the model is also stronger.