Research On Noise Reduction And Classification Of Heart Sound Signal

The prevalence and mortality of cardiovascular diseases are still rising,which not only seriously threatens the health of the people all over the world,but also greatly increases the economic burden of families and society.Heart sound signal can accurately reflect the physiological and pathological information of heart and cardiovascular,which is one of the important basis for clinical diagnosis of cardiovascular diseases.Accurate automatic heart sound analysis system can assist doctors in clinical diagnosis,improve the accuracy of diagnosis and make up for the lack of empirical judgment;At the same time,it can also be used for physical examination screening of large groups,reduce the cost of human and material resources of medical testing,and improve the efficiency of testing.Therefore,the study of effective heart sound analysis algorithm has important value and significance for the diagnosis of cardiovascular diseases.Based on the automatic heart sound analysis system,this paper studies the methods of heart sound noise reduction and heart sound classification.The main research contents can be summarized as follows:(1)In the research of heart sound noise reduction algorithm,this paper first analyzes the advantages and disadvantages of noise reduction methods based on wavelet analysis,empirical mode decomposition and optimal improved logarithmic spectrum amplitude,a heart sound noise reduction method based on heart sounds activity detection(HSAD)and improved minimum controlled recursive averaging(IMCRA)is proposed.In this method,the HSAD method is designed to judge whether the current heart sound frame is the fundamental heart sound frame,and the noise power in the fundamental heart sound frame and non-fundamental heart sound frame is dynamically estimated and updated by IMCRA algorithm and recursive smoothing algorithm respectively,so as to suppress the noise to the greatest extent.However,in the environment of low signal-to-noise ratio,HSAD method is prone to misjudge the fundamental heart sound and nonfundamental heart sound,and IMCRA method has noise underestimation in order to ensure that the heart sound signal is not distorted,which is easy to lead to noise residue.Therefore,based on HSAD and IMCRA noise reduction algorithms,this paper further introduces successive variational modal decomposition(SVMD)to reduce heart sound noise,which is mainly due to SVMD's ability to effectively separate high-frequency noise and improve the shortcomings of HSAD and IMCRA noise reduction algorithms in low signal-to-noise ratio environment.Experimental results verify the feasibility of the proposed algorithm.(2)Heart sound classification is the final means to realize automatic heart sound analysis.This paper mainly studies the classification of normal and abnormal heart sounds,and proposes a heart sound classification method based on power spectral density and convolution neural network.For the preprocessed data,the cardiac cycle is obtained through cyclic autocorrelation,and the timefrequency characteristics of cardiac cycle power spectral density with the same dimension are extracted by bilinear interpolation method,which are sent to convolutional neural network for training and testing.The experiment uses challenge 2016 data set for training and testing.The classification accuracy of the test set reaches 0.8472,and the sensitivity and specificity scores reach0.7763 and 0.9463.The overall performance is good.Compared with other algorithms,the results show that the algorithm obtains a higher overall score.