Research On Recognition Of CHD Heart Sound Signal Based On 1D-CNN And LSTM Algorithm

Cardiac auscultation is of great importance in the diagnosis of heart disease.Cardiac auscultation is currently the primary means of clinical diagnosis and screening for congenital heart disease.However,it often requires a wealth of clinical experience and is easily misdiagnosed by external environment and subjective factors.Therefore,it is very meaningful to study heart sound signals through digital signal processing techniques,especially in terms of deep learning algorithms.In this paper,the characteristic parameters of the heart sound signal extracted by Shannon energy are normalized as the feature vector,and the normal and congenital heart sound signals are identified by the one-dimensional convolutional neural network.The main research contents of this paper include:1.Pretreatment.For the unavoidable noise in the collected heart sound signal,the heart sound signal is de-noised by wavelet soft threshold method.2.Feature extraction.Aiming at the nonlinear characteristics of heart sound signal,the characteristic parameters of heart sound signal based on normalized Shannon energy method are proposed,and the format conversion and storage are carried out.Classification identification.Aiming at the one-dimensional characteristics of eigenvectors,an one-dimensional convolutional neural network(1D-CNN)method,an one-dimensional convolutional neural network(1D-CNN)with long-term and short-term memory network(LSTM)were proposed to combine normal and abnormal heart sound signals.Perform two-category identification.In this work,total 2307 cases of heart sound signal,including 2001 normal cases and 306 cases of pathological were used for training and recognition.Finally,the average recognition rates of 93.7% and 91.8% were got for 1D-CNN and 1D-CNN with LSTM respectively.The result shown that the above two networks have better effects on the classification and recognition of heart sound signals.