Study On MEMS Electronic Stethoscope And Heart Sound Feature Extraction For Early Noninvasive Diagnosis Of Coronary Heart Disease

Coronary heart disease(CHD)has been identified by the world health organization as the number one killer of human beings.In China,the mortality rate of CHD continues to increase year by year.Early detection for treatment is a necessary way to reduce the mortality of CHD.Changes in heart sounds and the appearance of murmurs are often the earliest signs of organic heart disease.Noninvasive diagnosis of CHD based on heart sound signal analysis and feature extraction has been widely concerned by scholars at home and abroad.Traditional stethoscope has been gradually replaced by electronic stethoscope since the electronic stethoscope has advantages of the real-time waveform storage and playback,easy to use,low cost,small size and so on.The signal-to-noise ratio(SNR)of electronic stethoscope is very important for the feature extraction of heart sound signal because even the very weak external noise may lead to the misjudgment of the pathophysiological information in the heart sound signal and then leads to the misjudgment of the disease diagnosis finally.In the case of complex noise types,the tradeoff between eliminating noise and retaining pathological information becomes very difficult.Therefore,electronic stethoscope with high SNR can greatly reduce the difficulty of data analysis and processing,improve the accuracy of data analysis results,and achieve twice the result with half the effort.Aiming at the practical application demand of early non-invasive diagnosis of CHD,an improved method of heart sound sensor’s structure was proposed to enhance the SNR of electronic stethoscope and a complete electronic stethoscope was fabricated.The data of heart sounds of CHD and other non-CHD were collected in the department of cardiology of the Second Affiliated Hospital of Shanxi Medical University with self-made MEMS electronic stethoscope.The differences between CHD and non-CHD heart sound characteristics were studied.The method of combining the first heart sound characteristic parameters with diastolic heart sound characteristic parameters to distinguish CHD from non-CHD is proposed and the accuracy of the diagnosis of CHD has been improved.The quantitative relationship between the degree of anterior descending coronary artery occlusion and the characteristic parameters of diastolic murmur was established.The whole research of this paper can be concluded as follows:1、Inspired by the principle of hydroacoustic signal detection,this paper proposes the microstructure design of heart sound sensor based on the microstructure vector hydrophone which imitate fishes’ lateral line.The sensor’s micro-structure has been fabricated by using Micro-Electro-Mechanical System(MEMS)technology and coupling encapsulated by choosing a kind of medical coupling agent as the filling material,which greatly reduced the signal attenuation caused by heart sound signal transmission to the body surface and further to the core sensitive unit.Finally,the performance of the proposed sensor is tested.The test results show the SNR of the proposed heart sound sensor is superior to 3200-type of 3M Littmann 8.2 db.2、In view of the characteristics of weak heart contractility and weakened amplitude of the first heart sound(S1)commonly existed in auscultation of heart sounds of chronic CHD,t A recognition method for CHD and non-CHD based on the first heart sound feature extraction is proposed.Firstly,the heart sound signal is preprocessed by filtering,resampling,normalization,and segmentation.Secondly,EWT is used to decompose S1 into several modal signals.The first two maximum points in the S1’s Fourier spectrum with a distance greater than 20 Hz are selected,and the nearest minimum points on both sides of the maximum points are found as the segmentation boundaries.Then S1 is decomposed into 5 modes and the instantaneous frequency(IF)of each mode is calculated by Hilbert transformation.Finally,kmeans clustering algorithm was used to cluster the IF of signals in different modes,and the IF of mitral valve(M1)and tricuspid valve(T1)in S1 were obtained.According to the three parameters of TD,Apeak_ratio and IF of T1,the decision tree classifier was designed.S1 was finally divided into four categories: normal S1,abnormal S1 split,S1 of CHD and S1 with abnormal amplitude.Although the amplitude of CHD’s S1 also has abnormal changes,a more significant feature of CHD’s S1 that distinguishes CHD from other non-CHD is the prominent decrease in the IF of T1.3、The quantitative relationship between the diastolic parameters of CHD and the degree of coronary artery occlusion is preliminarily established.This was studied from four different angles: the characteristic difference of diastolic heart sound between CHD and normal,the characteristic difference of diastolic heart sounds before and after stent implantation,the characteristic difference between diastolic turbulence murmur and diastolic valvular murmur and the characteristic difference of CHD with different degrees of left descending coronary artery occlusion.Fourier transform and spectrum segmentation were performed in the diastolic period lasting 128 ms starting from 100 ms after the end of S2 and the spectrum segmentation boundary was set as: [150 500] Hz.The spectrum distribution of the three modal signals is 0-150 hz,150-500 hz and > 500 Hz respectively.Through comparative study,it was found that the spectrum energy e(2)of the second diastolic mode of CHD was much higher than that of normal people,and e(2)significantly decreased after stent implementation.At the same time,e(2)increased gradually with the increase of coronary artery blockage in the anterior descending branch,except for acute myocardial infarction and severe coronary calcification.The characteristic parameter used to distinguish CHD’s diastolic turbulence murmurs from valvular heart disease’s diastolic murmurs is P3.Finally,the decision tree classifier was designed with the three characteristic parameters of e(2),P3 and T1,which improved the accuracy of identification of CHD and non-CHD.At the same time,according to e(2)and e(3),the degree of coronary artery blockage in the anterior descending branch was further subdivided into three types: about 30% blockage,about 50% blockage and more than 75% blockage.Finally,the purpose of early non-invasive diagnosis of CHD based on heart sound feature extraction is preliminarily realized.