Design And Implementation Of CHD Auxiliary Diagnosis System Based On Mobile Intelligent Terminal

Cardiac auscultation is the most simple and effective means to check the heart health,but it needs the auscultation doctor who has rich auscultation experience and excellent professional skills.The village doctors often lack this ability.In order to help village doctor to diagnose congenital heart disease（CHD）,a new kind of machine aided diagnosis method was studied by analyzing the heart sound using the digital signal processing and deep learning technology.An Android platform was designed using edge computing technology to realizes the of heart sound in this study.Aiming at the lack of heart sound analysis ability of the current mobile terminal,a auxiliary diagnosis system for CHD independent of cloud server is implemented on the mobile intelligent terminal based on Android platform.The system does not need network support,and can realize localization analysis of heart sound in remote mountainous areas without network or poor network environment.The following research works were done in this work.1.For the preprocessing process of heart sound running in mobile terminal,the algorithm of heart rate calculation and heart sound location is redesigned.The heart rate calculation relies on the simpler software envelope detection method,which can speed up the heart rate calculation and maintain high accuracy.In view of the need to locate the cardiac cycle completely in the current heart sound research,this thesis adopts S₁ location and fixed frame length segmentation method,which not only improves the speed of heart sound location,but also ensures the same start of each frame.2.Aiming at the low performance of mobile terminal,the heart sound classification network is designed as a lightweight dual input heart sound classification network.Under the premise of reducing the amount of model parameters and calculations,the network can ensure that the accuracy rate is basically the same or has a limited reduction.Simultaneously,for the input structure of the network,this thesis uses the mixed features of nonlinear and linear features.The nonlinear features are composed of Mel spectrum coefficients,Barker spectrum coefficients and their variants,which are mainly multi-dimensional feature maps.It can show the nonlinear characteristics of different frequency scales,and the feature information is more abundant.The linear feature is the spectrogram feature obtained by short-time Fourier transform.The function of the feature is to compensate the feature of the network when the number of training layers is deep.3.The algorithm and model mentioned above were ported to the Android platform in this work.The heart sound analysis algorithm is reconstructed by C language,which makes the algorithm has good portability and high efficiency.The system has the functions of heart sound collection,display,storage,heart rate calculation and analysis,and can obtain the heart sound classification results without the participation of cloud server.The classification accuracy of 92.89%and 89.63%were achieved in own data set and Physio Net/Cin C 2016 public data set respectively by using novel auxiliary diagnosis system.At the same time,the test results on different devices also show that the system has strong applicability and can realize the localization analysis of heart sound on mobile intelligent terminals of different manufacturers.