Delineation of Systolic and Diastolic Heart Murmurs via Wavelet Transform and Autoregressive Modeling

This paper describes a signal processing algorithm that utilizes wavelet transform and autoregressive modeling to identify heart sounds and to generate clinical features that characterize systolic and diastolic heart murmurs. A wavelet transform (WT) based on the Daubechies wavelet was adopted to facilitate the identification of the first and second heart sounds (S1 and S2) and to isolate systole and diastole periods without referring to the ECG waveform. Quantitative descriptors of heart murmur features such as the pitch frequency and configuration (crescendo, decrescendo, or plateau) were derived from either the systole or the diastole period where the murmur resides using a second order autoregressive (AR) whitening filters. The performance of this combined WT and AR modeling signal analysis was evaluated and demonstrated on selected systolic and diastolic murmurs and the results are presented in this paper.