Coherence analysis of heart sounds recorded from different auscultatory areas

A method based on cross-spectral and coherence analysis has been developed to compare the spectral details of heart sounds recorded from different auscultatory areas across the chest wall. The Vieira-Morf algorithm for multichannel AR spectral estimation has been used to estimate the coherence function between two channels of heart sound at a time. To reduce the “feed-across” problem associated with this algorithm, signals in different channels are partitioned in the frequency domain by using a bank of octave filters. This serves to restrict the dynamic range of the signals to be processed. The required parameters then can be estimated by using an AR model of lower order which In turn reduces the feed-across effect. The results of an experimental study indicate a coherence of more than 90% amongst the dominant frequency components in all channels and all frequency bands. The linear phase coherence observed amongst dominant frequency components, indicates a single delay in arrival of the intra-cardiac sounds to different auscultatory areas