Design Of Heart Sound Classification And Identification System Based On LabVIEW

Cardiovascular disease has brought new challenge to human beings and become aglobal public health problem. The heart’s and cardiovascular system’s mechanicalactivity can be reflected by heart sounds, whose detection is becoming one of theeffective methods of clinical assistant diagnosis for cardiovascular disease. Currently,there are two common ways to detect heart sounds in clinic, namely cardiac auscultationand phonocardiogram (PCG). But the cardiac auscultation technique is vulnerable to thesensitivity of human hearing and subjective experience for clinicians. Although PCGcan make up some deficiencies of cardiac auscultation, it also exist some shortcomings,which limit its application to some extent. Recently, with the gradual promotion of thecomputers and signal processing technology applications, design and development ofheart sound analyzer has become a trend in the field of heart sound analysis. Although,most of the heart sound analyzers can realize the simple function of time frequencyanalysis to heart sounds, they can hardly achieve classification and recognition of heartsounds. So the function of heart sound analyzer needs to be further improved. Giventhis, a heart sound classification and recognition system is designed based on theLabVIEW platform in this paper.First of all, the overall design scheme of the system is proposed in this paper, andthe system is divided into three subsystems which are preprocessing module, featureextraction module, modeling and recognition module. There are four modules in thepreprocessing module: denoising module, pre-emphasis module, framing module andendpoints detection module. The denoising module is designed by use of waveletdenoising method. Through experiment, the appropriate selection of the mother waveletfunction, threshold and decomposition scale are determined, respectively. Thepre-emphasis module is implemented by a first-order digital filter. Hamming window isused to design the framing module. The endpoints detection module is finished based onthe principles of short-time energy and short-time zero crossing rate.Secondly, in the feature extraction module, this paper introduces the principles onthe extraction of Mel frequency cepstrum coefficient and its four improved parameterswhich are Mel frequency cepstrum coefficient combined with its first-order differentialcoefficient and Delta feature, respectively, discrete wavelet transform Mel frequencycepstrum coefficients and its first-order differential coefficient (DWPTMFCC+ΔDWP TMFCC). At the same time, on the LabVIEW platform, the difficulty of extracting theabove five parameters is analyzed and the corresponding solutions are presented.And lastly, in the modeling and recognition module, the classical Gaussian MixtureModel (GMM) is employed to classify and recognize the heart sounds. This paperintroduces the principle of GMM, and analyzes the drawbacks of K-means algorithmacted as the traditional parameters initialization algorithm in GMM. To solve theproblems, three modified algorithms are proposed, namely, Approximate FuzzyC-means Clustering algorithm, Weighted Fuzzy C-means Clustering algorithm andWeighted Optional Fuzzy C-means algorithm (WOFCM). The key points of thismodule are the design and implementation of the four recognition models’ training andrecognition process on the LabVIEW platform.The final overall system is completed by building the above modules together.From the angle of feature parameters and recognition models, the system are tested, inother words, the designed system in this paper is used to classify the heart soundsacquired from clinic which include the normal heart sounds and nine kinds of abnormalsignals, namely mitral stenosis, aortic stenosis, aortic insufficiency, ventricular septaldefect, pulmonary stenosis, arrhythmia, mitral insufficiency, splitting of first sound andsplitting of second sound. Considering the recognition rate and time, the systemperforms the best recognition performance when the DWPTMFCC+ΔDWPTMFCC andimproved GMM using WOFCM are served as the feature parameter and recognitionmodel, respectively, and the results of abnormal signals are more effective improvement.Therefore, the original whole system is simplified by using this parameter and model,and then the final system of heart sound classification identification are obtained, whichcan complement the function of heart sounds analyzer and has potential application inthe research of the cardiac activity and the diagnosis of cardiovascular diseases.