Design And Application Of Electronic Stethoscope

Cardiovascular disease has become one of the greatest threats to human health in the world.As a qualitative method,the result of artificial auscultation depends entirely on the doctor’s clinical experience.Computer-aided detection technology can obtain quantified characteristic parameters by measuring the tone,intensity and frequency components of heart sounds,which will help diagnose cardiovascular diseases more accurately.Computer-aided intelligent auscultation technology mainly involves three parts: heart sound noise reduction processing,location segmentation,feature extraction and classification.Although researchers have conducted in-depth research at one of these points,there still lacked a complete processing system.In view of the above problems,the thesis attempt to develop a complete solution.The work of research subject focuses on two parts including hardware and algorithm.A portable and low-power electronic stethoscope was designed in terms of the preset scheme,the thesis gives a detailed explanation of the overall hardware design framework and the design details of each module,and briefly introduces the internal software ideas.In terms of algorithm,this project has optimized the three parts of the heart sound processing flow: noise reduction,location segmentation,feature extraction and classification.Among them,in the heart sound noise reduction stage,we adopts an optimization algorithm based on empirical mode decomposition and reconstruction.By setting a threshold for each empirical mode component after decomposition,we compare its actual energy density value with the estimated energy density value.Great heart sound noise reduction effect was achieved after the heart sound signal is reconstructed.In terms of heart sound localization and segmentation,The accurate positioning of the heart sound components is realized by calculating the energy entropy of the heart sound signal based on the S transform to achieve pre-positioning,and then implementing an improved window width optimization algorithm based on the S transform;In the feature extraction and classification stage,this thesis screens the manually extracted features,and uses the convolutional neural network to extract the model feature extraction.The two extracted features are combined and inputted into the sequence classification model based on the bidirectional gated recurrent unit.The model was trained by a large tagged dataset of heart sound and its classification accuracy was stable to 95.3%.Based on the prototype of the electronic stethoscope,400 sets of normal and abnormal heart sound data were collected in this thesis.The trained model was used for disease simulation diagnosis.After preprocessing the collected heart sounds,the classification accuracy of the final model reaches 95%,which is better than other similar models.This research project has an important significance for the pre-screening of family heart sound diseases and the hospitalassisted diagnosis.