Research On Heart Sound Recognition Of Congenital Heart Disease Based On Bi-LSTM Network

Heart sounds can truly reflect the working status of the heart,which is important for doctors to diagnose cardiovascular diseases.The clinical diagnosis of congenital heart disease is divided into two stages,initial diagnosis and definite diagnosis.In the initial diagnosis or screening stage,as a clinical examination tool,stethoscope is mainly used to determine suspected patients through cardiac auscultation,which requires doctors to have rich clinical experience.Usually,the grassroots doctors are often not competent.Misdiagnose or diagnose delay may happen if they are not experienced enough.At the stage of definite diagnosis,echocardiography is mainly used for reexamination of suspected patients.Due to the high price of this equipment,it cannot be equipped to the township hospitals and other rural hospitals.At present,the screening of congenital heart disease is mainly based on cardiac auscultation.Due to the unbalanced regional development in China,especially the lack of grassroots medical resources in Yunnan,the screening of congenital heart disease is basically completed by provincial medical teams traveling to the countryside,which is very unfavorable for the timely and early detection of patients with congenital heart disease.Therefore,it is particularly important to analyze and study the heart sound signal of congenital heart disease,extract relevant pathological features,and study the machine-assisted diagnosis technology.It can improve the accuracy rate of screening of congenital heart disease for primary doctors.A new kind of heart sound signal recognition algorithm based on Bi-LSTM network and Mel frequency cepstrum coefficient method was put forward in this work,which solveed the classification problem of heart sound signals and provided a reference for the clinical diagnosis of congenital heart disease.The analysis and processing of heart sound usually involves signal preprocessing,feature extraction,and classification recognition three steps.In the signal preprocessing part,noise reduction of heart sound.envelope extraction,and others were included.The original heart sound signals were subjected to wavelet 5 layers.The noise reduction processing was done by recombination of some wavelet layers to obtain de-noised signal.Then the envelopes of the heart sound signal were extracted by using Hilbert method.Finally,6 seconds length of heart sound signal was intercepted with S1 as a starting point.In the feature extraction part,three feature extraction methods of LPCC,BFCC,and MFCC were used to extract the Mel cepstrum frequency coefficient as features of congenital heart disease.Finally,BP,RNN,LIST and Bi-LIST neural networks were applied to make classification and identification of congenital heart disease.Comparing the above classification methods,the Bi-LSTM network had a better identification ability.The results showed that the Bi-LSTM network had a good recognition effect.Its correct recognition rate for the normal and abnormal heart sound signal reached 84.2%.