Detection of the intensity of heart murmurs : intensity identification

• Main Author: Luo, Guoliang
• Other Authors: Soh Cheong Boon
• Format: Final Year Project
• Language: English
• Published: 2015
• Subjects: DRNTU::Engineering::Electrical and electronic engineering::Control and instrumentation
• Online Access: http://hdl.handle.net/10356/61989
• Institution: Nanyang Technological University

This report will present a methodology and algorithm to detect heart murmurs from raw phonocardiogram signal (PCG). The algorithm is divided into 4 phases namely: ‘Preprocessing’, S1/S2 detection, Murmur detection and Murmur intensity calculation. In the ‘Preprocessing stage, the raw PCG signal is filtered with a simple low pass filter to remove the high frequency noise. By applying fast fourier transform, the range of a normal heart sound (HS) for the low pass filter is known immediately. Normalization of the signal is done to reduce the dependence of the segmentation method on the offset and the peak-to-peak magnitude of the signal. In the 2nd stage of the algorithm, wavelet transform is applied to the signal using the 4th order Daubaucies wavelet as it provides the smallest time-frequency localisation. Wavelet transform theory is commonly used by researchers in PCG analysis as heart sounds are non-stationary signals and it is effective in analysing heart sounds with its time-frequency localisation property. After the signal is filtered using wavelet transform, the output is squared and a moving window integration is applied to smoothen the signal and amplify the S1/S2 peaks. The start and end of S1 and S2 are then automatically detected and plotted. In the 3rd stage of the algorithm, murmurs are detected by decomposing the signal using Wavelet Transform into 5 levels of detailed and approximate coefficients. Then, murmurs are identified by the location of the occurrence of the murmurs with respect to S1 and S2. Finally, in the last stage of the algorithm, murmur intensity is calculated by determining the S1/S2 locations at the level 5 detail of the PCG signal. Then, the relevant points (depending on the type of murmur present in the PCG signal) are scaled to represent them in the original PCG signal where the points between the relevant points are squared and summed to determine the murmur intensity.