Research On Heart Murmur Of Hypertrophic Cardiomyopathy Based On Feature Extraction And Optimization Algorithm

At present,topics such as sub-health and population aging have received widespread attention.Cardiovascular disease is a chronic disease that seriously affects human healthy life.Its prevention and diagnosis have always been one of the important issues that the medical community is facing urgently to solve.The analysis of heart sound has an important impact on the diagnosis of cardiovascular diseases.Aiming at the problem that the classification accuracy of hypertrophic cardiomyopathy is not high,this thesis studies related algorithms from the perspective of feature parameter extraction and optimization.The research content mainly includes the following aspects:(1)Feature extraction in wavelet domain.Based on the wavelet transformation,the single sub-band reconstructed signals with different frequency components are obtained,and the normalized energy features of the reconstructed signal are extracted.The single sub-band energy features represent the energy ratio of the different frequency components of the signal in the original signal.From a statistical point of view,the difference between the energy features of normal and hypertrophic cardiomyopathy heart sounds is analyzed,and a preliminary judgment is made on the frequency band of the heart murmur.(2)The dimensionality reduction and combination analysis of heart murmur features based on eigen decomposition.Method one uses principal component analysis to locate the frequency band of the heart murmur components,extracts the corresponding quantitative indicators,and combines them with the degree of difference to obtain the combined heart murmur features.Method two uses linear discriminant analysis to linearly combine the original features to obtain the combined features.The results of clinical heart sound classification show that the classification effect of the quantitative index of heart murmur extracted by method one is better than the classification effect of the index extracted by method two,which can achieve the purpose of reducing feature redundancy and improving classification accuracy.(3)Optimization and combination of feature subsets of heart murmur based on feature selection method.The article uses three feature selection methods of chi-square test,binary particle swarm optimization(BPSO),and extremely random tree to extract the effective feature subsets of heart sound signals,and combine them according to the degree of difference.The clinical heart sound data verification results show that the "BPSO-based feature subset selection and combination method" obtains a better classification accuracy rate of 95.86%,and the number of features used is 23.08% of the original features.The results show that the method can achieve the purpose of screening out effective feature subsets,improving classification accuracy and reducing classification complexity.The method automatically outputs classification results with a high degree of automation,and conforms to the basic concept of intelligent diagnosis of cardiovascular diseases.(4)Heart murmur time-frequency domain feature combination optimization analysis.The article introduces a feature combination method based on time-domain systolic diastolic energy features and frequency domain scaling factor to extract the time-frequency domain combination features of heart murmurs.The optimal classification accuracy of the combined features can reach 95.37%,which is better than the result of directly using the original time domain or frequency domain features for classification.The method makes full use of the pathological characteristics of abnormal heart sounds and quantifies them by using feature combinations to achieve the purpose of improving the classification accuracy of normal and hypertrophic cardiomyopathy heart sounds.