Heart Sound Signal Analysis Method Based On EMD Adaptive Reconstruction

Heart disease is regarded as one of the most serious threats to human health.As a physiological signal,heart sound can effectively reflect the specific information contained in cardiovascular disease.The analysis of heart sound signals can not only provide doctors with relevant reference in clinical diagnosis,but also has the advantages of non-invasiveness,low cost,and convenient detection compared with an electrocardiogram.The heart sound signal has the characteristics of non-linearity and complexity,and different heart diseases have their features and discipline.To realize the effective diagnosis and classification of a variety of heart sound,this paper starts with the analysis of heart sound of heart valve defect diseases and congenital heart disease.The analysis of non-linear and non-stationary signals and quantitatively extracts features and selects feature parameters from the time domain,frequency domain,and non-linear space.Through classification and recognition,auxiliary diagnosis of cardiovascular disease is achieved.The research contents include:(1)This paper studies an analysis method based on EMD adaptive reconstruction of heart sound signals.Firstly,the original heart sound signal is transformed by empirical mode decomposition(EMD)to obtain the corresponding intrinsic modal components(IMFs).Secondly,the correlation coefficient(Corr)and root mean square error(RMSE)of the IMF component signal and the original heart sound signal is calculated,analyze the adequate information and noise distribution of the original signal contained therein.An adaptive threshold evaluation index is then proposed to select the IMF component signals that meet the conditions.Finally,the sub-signal of heart sound is reconstructed.Experiments show that the proposed algorithm can reduce noise without heart sound segmentation and provide better input signals for subsequent feature extraction,feature selection,and classification.(2)This paper proposes an improved adaptive reconstruction method based on Hausdorff distance.After EMD transformation,the Hausdorff distance(HD value)between IMFs and the original heart sound signal is calculated.According to the proposed adaptive threshold based on Hausdorff distance,practical IMF components are selected to reconstruct the heart sound signal.Finally,the results of Corr and RMSE are compared.According to the index’s result,it is further verified that the method had better noise reduction and had more apparent characteristic information in reconstructed heart sound signals.(3)After preprocessing and analysis,the multi-dimensional features of the reconstructed heart sound are extracted.The feature selection method is then combined with the machine learning classification method to verify the effectiveness of the improved adaptive reconstruction method based on Hausdorff distance in this paper.Firstly,this paper extracts 40 feature parameters based on the time domain,frequency domain,and nonlinear space.Secondly,using 6 feature screening and sorting algorithms,the feature parameter sets are sorted in descending order of feature importance scores.Three machine learning methods are then selected and input the feature sets of different dimensions into the classifier.Finally,determine the best feature dimension with the classification accuracy.Finally,the preprocessing method based on EMD adaptive reconstruction in this paper is combined with the methods of multidimensional feature extraction,feature screening and classification,and the best analysis and processing method for heart sound signals is determined through simulation experiments.The results show that the feature extraction,screening and classification method based on EMD adaptive reconstruction of heart sound signals in this paper can effectively improve the classification accuracy of multiple types of cardiovascular diseases,and it has practical application significance.