Research On Automatic Classification Algorithm Of ECG Beats Based On Cross-Wavelet Transform

The morbidity and mortality of cardiovascular diseases such as hypertension,coronary heart disease and myocardial infarction are increasing year by year.They have become the "first killer" of human health.How to diagnose cardiovascular diseases in time has become an urgent problem.As a non-invasive measurement method,ECG signal is the main diagnostic basis for clinicians to judge cardiovascular diseases.However,the huge amount of data of ECG signals requires a lot of time for doctors to interpret and analyze.Therefore,the study of automatic analysis algorithm of ECG signals can free doctors from heavy manual work,reduce the subjective differences among doctors,and thus improve the efficiency of diagnosis of cardiovascular diseases.In this paper,the wavelet coherence spectrum and cross spectrum of heartbeat are established by cross-wavelet transform.Twenty groups of features related to heartbeat classification are extracted from the coherence spectrum and cross spectrum.Then,support vector machine classification method is introduced to classify the extracted features.The optimization method of optimal parameters is studied in the classification,which shortens the optimization time.Aiming at the problem that the classification accuracy is greatly affected by reference beats,an automatic generation algorithm of reference beats template based on point set grouping registration is proposed,which optimizes the selection of reference beats and realizes the automatic classification of normal beats,left bundle branch block,right bundle branch block,pacemaker heartbeat,ventricular premature contraction and atrial premature contraction.Specific research contents and contributions include:(1)A method based on cross-wavelet transform is proposed to represent the chara cteristics of heart beat signals.The wavelet coherence map and cross-plot are establish ed to show the time-frequency characteristics in the form of atlas,and the gray level co-occurrence matrix is established in four directions(?=0?45,?90,?135,?).After calcula ting and extracting 20 sets of features,such as energy,entropy and moment of inertia,etc.(2)An improved grid search algorithm is proposed for automatic heart beat classification.The penalty coefficients C and kernels?are obtained by rough searching with a larger step distance.The optimal parameters are obtained by narrowing the search step and fine searching near the two parameters.The classification accuracy of heartbeat reaches 94.96% and the classification time is shortened.(3)Aiming at the problem that the classification accuracy of heartbeat is greatly affected by reference heartbeat,an automatic generation algorithm of reference heartbeat template based on point set grouping registration is proposed,which converts the generation of heartbeat template into the registration of point set grouping.CDF-HC and Holderother methods are comprehensively analyzed.Considering the fitting degree and registration time of the template,a point set grouping registration method based on information potential is proposed to generate reference heartbeat template,which improves the accuracy of heartbeat classification to 96.17%.