Reserch On Recognition Of CHD Heart Sound Based On Wavelet Cepstrum Coefficient And Probabilistic Neural Network

Cardiovascular diseases and cerebrovascular diseases have high morbidity and mortality,and are difficult to prevent and treat.Cardiovascular diseases are one of the common cardiovascular and cerebrovascular diseases.Heart signal is an important physiological signal.Heart auscultation is the main means of initial diagnosis of congenital heart disease.Abnormal information can be found before heart disease changes,so as to prevent the occurrence of heart disease.At present,the clinical method of obtaining cardiac information is doctor auscultation,but relying solely on doctor auscultation will lead to inaccurate judgment and inevitably lead to misdiagnosis.Therefore,the analysis of feature extraction and recognition of heart sound signals is beneficial to the diagnosis of congenital heart disease,and provides objective basis for the clinical diagnosis of heart disease.This paper explores the feature extraction algorithm of wavelet cepstrum coefficients and probabilistic neural network for the analysis and recognition of heart sound signals of congenital heart disease.The main work is divided into three parts:signal preprocessing,feature extraction and classification recognition.Signal pre-processing:Using digital filter and wavelet transform to denoise,after comparing the results,this paper chooses wavelet transform to remove heart sound noise.Comparing the advantages and disadvantages of different envelope extraction methods,using Viola integral method to get the smooth envelope of heart sound signal;then,using double threshold segmentation method to get the heart sound signal’s cardiac cycle;feature extraction:calculating the wavelet cepstrum coefficients of each cardiac cycle signal,and extracting the wavelet cepstrum coefficients as the feature vectors of the signal;The probabilistic neural network is used as the signal recognition network,and the extracted wavelet cepstrum coefficients are used as the eigenvectors to input the probabilistic neural network for recognition.In this paper,2083 heart sounds samples were randomly selected from the heart sounds database constructed by our research group(including 1700 normal heart sounds,383 congenital heart disease heart sounds,80%training set and 20%test set).The experimental results show that the recognition rate of this method is 91%specificity,86.7%sensitivity,and 90.2%accuracy.The experimental results show that this method can effectively recognize the heart sound signal of congenital heart disease.