1 Application of Hilbert Huang Transform on Phonocardiogram Signals Ashish Raj Department of Electrical and Electronics Engineering RV College Of Engineering, Bangalore, KA, 560059 India e-mail: (ashishraj.ee16@rvce.edu.in). Abstract—Coronary Diseases are a broad range of chronic heart related problems, the major arteries become too narrow. The coronary arteries supply oxygen and blood to the heart. A major cause of Coronary Disease is build up of cholesterol in the arterial walls, creating plaques. The plaque deposition increase the speed of blood flow because of effective decrease in cross sectional area of the blood vessels, this is an example of the principle of continuity in fluid dynamics. This increased blood pressure produces a turbulence which affects the systolic and diastolic murmur that is recorded by the phonocardiogram. The cardiac sound signals are classified as non linear and non stationary because they cannot be modelled after and because of their non linearity and non stationary nature, other signal processing tools like Fast Fourier and wavelet transforms cannot be used. Hilbert Huang Transform is a tool specifically used to analyse such signals. An implementation of HHT is done using MATLAB, This involves developing multiple functions including an Empirical Mode Decomposition function. The tools used for signal processing involved the use of MATLAB DSP Toolbox. The data required that is Phonocardiogram signals from patients was obtained from Physionet.org where the data is available for use for academic and research purposes. the cardiac sounds from various subjects are analyzed. The initial results show that coronary diseases can be detected by using Hilbert Huang Transform on cardiac sound signal. Index Terms—Hilbert Huang Transform, Phonocardiogram, Coronary Disease. I. I NTRODUCTION C ORONARY Diseases are a major type of heart disease in the world and it develops silently over the years without showing any symptoms. Early diagnosis of these kind of diseases has gained utmost importance in medical research. Heart sound signals provide clinicians with valuable diagnostic and prognostic information about the functioning of arterial valves and vessels.[1] Some evolving therapies, including new ACE inhibitor drugs have been developed in clinical practice. Nevertheless, outcome is still unsatisfactory. Advanced signal processing methods such as Short Time Fourier Transform and Wavelet Transform etc exist with some limitations. Short Time Fourier Transform involves an intrinsic trade-off between time resolution and frequency resolution. The wavelet transform has received considerable attention in recently years. It provides a multi- resolution representation of signals, whereas, it is not adaptive in nature; once the wavelet mother function is given, one will have to use it to analyze all the data. In addition, wavelet transform also underlies an uncertainty principle.[2][3] Hilbert Huang Transform is a proven tool to analyze nonstationary and nonlinear signal. The main steps in HHT are the Empirical Mode Decomposition (EMD) and Hilbert transform. EMD decomposes the cardiac sound signals into a finite number of Intrinsic Mode Functions (IMFs). Hilbert transform of IMFs can yield instantaneous frequency and instantaneous amplitude. The local energy and instantaneous frequency derived from the IMFs give the fine-resolution frequency- time distribution of the energy that is designated as Hilbert spectrum. The three-dimensional distribution can reflect the inherent essential characteristic of the signal. In this paper, we present a method to analyze the cardiac sound signals based on Hilbert Huang Transform. The paper is organized as follows: section 2 introduces HHT. In section 3, Implementation of HHT on MATLAB is discussed; Results and discussions are in section 4. Finally, some conclusions are given in section 5. II. HHT The implementation of Hilbert Huang Transform on MATLAB follows the shown flowchart. The HHT Function calls the Extrinsic Mode Decomposition function first. The EMD function consists of the Finding peaks functions which finds all the extremas in the signal[12]. Then the spline function is used to fit a cubic spline passing through all the extremas of the signal. The EMD function then calls the IMF and IMF check functions. In the figure 3.1, A flowchart of HHT function and EMD function as implemented in the work is shown.