Comparison of binomial, ZAM and minimum cross-entropy time-frequency distributions of intracardiac heart sounds

One reason why automated heart sound analysis remains unfeasible may be the inadequacy of conventional spectral techniques in representing the nonstationary, multicomponent characteristics of phonocardiograms. This study compares several generalized time-frequency distributions (GTFDs) applied to intracardiac phonocardiograms (ICP) obtained at rest in six patients using catheter-mounted transducers. ICPs were bandpass filtered (50-500 Hz) and digitized at 4 kHz. The TFDs employed in this study were the binomial transform (BT), Zhao-Atlas-Marks (ZAM) distribution, minimum cross-entropy (MCE) distributions and the spectrogram. The BT and MCE representations in particular show improved concentration of ICP energy in time-frequency vs. the spectrogram. These findings suggest that several GTFDs may prove useful in the design of automated auscultation systems.<
>