Refined analysis of heart sound based on Hilbert-Huang transform

Abstract

Heart sound signal, which reflects the conditions of human heart, has more advantages than ECG in some ways. Nowadays, the computer-aided diagnosis of heart sounds becomes available. In this research, telemedical auscultation and computer-aided diagnosis of heart sounds are combined. This paper analyzed heart sounds based on Hilbert-Huang transform from the time domain and frequency domain, then extracts a series of parameters which are useful for computer-aided diagnosis. Firstly, heart sounds are preprocessed by wavelet transform and Huang-transform technologies. Wavelet threshold denoising can effectively remove the noise of heart sounds, while Huang transform can extract a series of intrinsic mode functions, and choose the appropriate intrinsic modal functions, which can effectively remove the low-frequency noise of signal. This paper also extracts the envelopes of heart sounds based on Hilbert transform. The heart sounds are segmented effectively by Hilbert envelops, thus time domain features of the heart sound can be extracted more accurately. Finally, pathologic heart sounds were analyzed by using Hilbert-Huang transform. Heart sounds are collected in Shengjing Hospital of Chinese Medical University with the designed telemedicine consulting system for auscultation. Heart sounds of patients with acute myocardial infarction (AMI), and those of patients with coronary artery disease (CAD) are collected. The Hilbert-Huang transform theory, instantaneous frequency, Hilbert spectrum and its extracted boundary spectrum were used in this paper. Due to the good time-frequency resolution of Hilbert-Huang transform, five characteristic parameters are defined. Through comparative analysis of the three classes' heart sounds, the results demonstrated that the five features can differentiate the three classes effectively at the accuracy of 80%.