Classification of Heart Sounds Associated With Murmur for Automatic Diagnosis of Cardiac Valve Disorders

Abstract

The heart is one of the vital organs and cardiovascular diseases have been a major cause of deaths in the world. The heart sound is till the primary tool for screening and diagnosing many pathological conditions of the human heart. Abnormal heart sound is one of the precursors of many serious heart diseases; heart failure, coronary artery disease, hypertension, cardiomyopathy, valve defects, arrhythmia. This study concerns only heart valve defects. Cardiac auscultation is act of listening to heart sounds. Any abnormality in the heart sound indicates some problem in the heart. The abnormality in the heart sounds start appearing much earlier than the symptoms of the disease start showing. In this study, the PCG signal i.e. the digital recording of the heart sounds has been studied and classified into three classes, namely normal signal, murmur signal and extra sound signal. This study focuses on denoising of Phonocardiography signals using selected wavelet families, show superior signal denoising performance due to their properties of multi-resolution, including thresholding techniques, and signal decomposition levels. A total of 15 features from different domain have been extracted and then reduced to 7 features. The features have been selected on the basis of correlation-based feature selection technique. The selected features are used to classify the signal into the pre-defined classes using multiclass SVM classifier. The performance of our denoising algorithm is evaluated using the signal to noise ratio, percentage root means square difference, and root mean square error. The experiment shows that the level of decomposition, types of wavelets and thresholding techniques are the most important parameters affecting the efficiency of the denoising algorithm. Better SNR values compared with references revealing that the 4th level of decomposition is the optimal level for signal decomposition. The performance of the classification result is evaluated using the parameters accuracy, specificity, and sensitivity. An accuracy of 97.96% is achieved using multiclass SVM classifier. Finally, comparison was done with other related studies to optimize the performance of the proposed algorithm.