Research On Intelligent Prediction Methods Of Sudden Cardiac Arrest

With the rapid development of economy,the intense social rhythm and irregular living habits have caused the number of sudden deaths to increase year by year.Cardiac arrest is one of the main causes of sudden death.The key to improve the survival rate of cardiac arrest patients is timely rescue.According to the survey,the recovery rate of cardiac arrest rescue in China is less than 2%,which is closely related to the short-term and unpredictable characteristics of cardiac arrest.Thus,cardiac arrest prediction can help to solve the problem.In order to improve the accuracy of cardiac arrest prediction,this dissertation focus on the research of intelligent prediction methods of sudden cardiac arrest.Firstly,in order to extract the heart beats for machine learning from ECG,a QRS waveform detection algorithm using slope difference and adaptive threshold is proposed.The algorithm first calculates the slope difference,then the slope difference is compared with thresholds to judge whether it is a QRS wave.Once a QRS wave is detected,the thresholds are updated according to current peak.The algorithm has good real-time performance and precision,which is suitable for the extraction of heart beats.Then,considering the poor precision of current ECG features,we propose a cardiac arrest prediction method based on wavelet transform and SVM.The heart beats are decomposed with wavelet firstly.Then the decomposed low frequency coefficients are extracted,and their dimensions are reduced using PCA.After that,the labels of low-dimensional vectors are obtained using SVM classifier.The hyperparameters of the SVM are determined by a grid search method combined with cross-validation.This method greatly improves prediction precision and sensitivity at the prediction pernod of 5 minutes.Finally,considering that existing algorithms are difficult to make early prediction before cardiac arrest,we propose a cardiac arrest prediction method based on CNN.The method constructs a 1D CNN based on general 2D CNN,including three convolutional layers,three pooling layers and a fully connected layer.This method can make effective prediction at the prediction period of 30 minutes,and has great clinical value.All in all,this dissertation proposes effective cardiac arrest prediction methods,which are superior to the existing methods in the accuracy and prediction period.At the same time,some suggestions for future are illustrated,providing ideas for further research of the subj ect.