Research On Classification Method Of ECG Signals Based On Wavelet Packet And Neural Network

Electrocardiogram(ECG)is the most direct reflection of heart electrical activity in the human body and it is one of the important bases for doctors to diagnose and treat heart diseases.With the progress of science and technology,automatic analysis and diagnosis techniques based on ECG have been extensively investigated to detect and diagnose cardiac diseases.Automatic analysis and diagnosis techniques based on ECG can not only greatly decrease the workloads of doctors but also significantly improve the efficiency and accuracy of ECG classification,which have great practical application values in the diagnosis and treatment of cardiac diseases.Therefore,this thesis is mainly focused on ECG classification methods in automatic analysis and diagnosis techniques.The main contents of this thesis include ECG signals feature extraction and classification.In order to extract stable ECG features effectively,this thesis presents an algorithm based on wavelet packet decomposition(WPD)combined with the statistical analysis for ECG signals feature extraction.Thereafter ECG signals are decomposed into four levels by using WPD.The singular value,the standard deviation and the maximum value of the 16 wavelet packet coefficients(WPCs)in fourth level are separately calculated based on the statistical method.A total of 48 dimensions statistical features of WPCs are obtained as the ECG feature space.To improve the efficiency and accuracy of ECG signals classification as much as possible,a novel algorithm based on genetic algorithm-back propagation neural networks(GA-BPNN)for feature selection and classification is proposed in this thesis.The statistical feature space of WPCs is processed by GA for dimension reduction.A total of 25 dimensional features containing more representative ECG information are filtered.Meanwhile the weights and biases of BPNN are optimized by GA.The filtered features are placed into the optimal BPNN classifier for ECG recognition.In this thesis,the ECG signals derived from the MIT-BIH arrhythmia database are classified into six categories,namely,normal beat(N),left bundle branch block beat(L),right bundle branch block beat(R),atrial premature beat(A),paced beat(P),and premature ventricular contraction(V).The presented method with MIT-BIH arrhythmia database yielded the classification accuracy of 97.78%,a sensitivity of 97.86%,a specificity of 99.54%and a positive predictive value of 97.81%.Moreover,an ECG signal acquisition experimental platform is constructed to collect six types of ECG signals to demonstrate the utility of the proposed method.The experimental results obtained with the established acquisition platform illustrated that the GA-BPNN method achieved 99.33%accuracy,99.87%sensitivity,99.36%specificity and 99.33%positive predictability.Experimental results show that the proposed algorithms for feature extraction and classification can efficiently extract and filter the most representative features effectively,and the high classification accuracies of six types of ECG signals can be achieved by using the optimal classifier.Therefore,the presented algorithms based on WPD and neural networks in this thesis can be effectively applied in the arrthythmia classification,which has important sense for the detection,diagnosis and treatment of heart diseases.