Classification of heart disorders based on tunable-q wavelet transform of cardiac sound signals

Abstract

The mechanical action of the heart generates sounds which can provide diagnostic information about the functioning of the cardiovascular system. Cardiac auscultation is an important means to diagnose heart disorders by listening to the heart sounds using conventional stethoscope. The traditional cardiac auscultation techniques require sophisticated interpretive skills in diagnosis and it requires long time to expertise. The heart sounds often last for a short period of time and pathological splitting of the heart sound is difficult to discern using traditional auscultation because human ears lack desired sensitivity towards heart sounds and murmurs. Therefore, the automatic heart sound analysis using advanced signal processing techniques based on digital acquisition of these sounds can play an important role. The heart sounds can be captured and processed in the form of cardiac sound signals by placing an electronic stethoscope at the appropriate location on the subject’s chest. The cardiac sound signals can be used to extract valuable diagnostic features for detection and identification of the heart valve and other disorders. In this book chapter, a new method for segmentation and classification of cardiac sound signals using tunable-Q wavelet transform (TQWT) has been proposed. The proposed method uses constrained TQWT based segmentation of cardiac sound signals into heart beat cycles. The features obtained from heart beat cycles of separately reconstructed heart sounds and murmur can better represent the various types of cardiac sound signals than that of containing both. Even the parameters evolved during constrained TQWT based separation of heart sounds and murmur can serve as valuable diagnostic features. Therefore, various entropy measures namely timedomain based Shannon entropy, frequency-domain based spectral entropy, and nonlinear method based approximate entropy and Lempel-Ziv complexity have been computed for each segmented heart beat cycles. Two features have been created by the parameters that have been optimized while constrained TQWT namely the redundancy and the number of levels of decomposition. These ten features form the final feature set for subsequent classification of cardiac sound signals using artificial neural network (ANN) based technique. In this study, the following classes of cardiac sound signals have been used: normal, aortic stenosis, aortic regurgitation, splitting of S2, mitral regurgitation and mitral stenosis. The performance of the proposed method has been validated with publicly available datasets. The proposed method has provided significant performance in segmentation and classification of cardiac sound signals.