Classification of heart sounds using discrete time-frequency energy feature based on S transform and the wavelet threshold denoising

In view of the important role of heart sound signal in the prevention and diagnosis of cardiovascular diseases, this paper focused on the preprocessing and classification of heart sound signal. Firstly, since the signal-to-noise ratio cannot accurately reflect the denoising effect when the original signal is unknown, the noise envelope area as an evaluation index of the denoising effect for heart sound signal was proposed, and then the parameters of the wavelet threshold denoising method were optimized based on it. Next, the heart sound signal was denoised by the wavelet threshold denoising method with the optimal parameters. Secondly, the heart sound signal envelope was extracted based on the normalized average Shannon energy, and the double threshold method was improved by introducing the extra heart sounds culling function. Afterwards, the heart sound signal with murmur or extra heart sounds was located and segmented by the improved double threshold method. Finally, a discrete time-frequency energy feature based on S transform was proposed, and then the heart sound signals were classified by the Support Vector Machine based on it. The test results show that the discrete time-frequency energy feature based on S transform can better reflect the difference between different types of heart sound signals and has better classification effect compared with the conventional features. (C) 2019 Published by Elsevier Ltd.