Heart murmur detection and configuration estimation using wavelet transform and linear fit

This paper describes a signal processing procedure that identifies the first and second heart sounds (Si and S2) from repeated cardiac cycles; separates the systole from the diastole; isolates suspected murmurs found within; and delineates murmur configuration in regards to its shape. The identification of heart sounds was facilitated by performing the wavelet transform (WT) based decomposition and approximation using Daubechies (db6) wavelet, and Si and S₂ were correctly identified without referring to the ECG waveform. We extended our previous studies on heart murmur features extraction and emphasized in this study on delineating heart murmur configuration either as crescendo, decrescendo, crescendo-decrescendo, or plateau. The said task was accomplished by first segmenting and computing the energy index measure and the averaged absolute amplitude of the suspected murmur, and then estimating the slope(s) of the averaged murmur signal energy and the averaged absolute amplitude through a linear fit. The procedure outlined was evaluated using both emulated systolic murmurs and clinically recorded heart murmurs. The evaluation has shown satisfactory and promising performance for future expansion.