Comparison of short-time Fourier, wavelet and time-domain analyses of intracardiac sounds

Heart sounds provide clinicians with valuable diagnostic and prognostic information. They are repetitive in nature, but reflect complex mechano-acoustical events which have been inadequately described by traditional digital signal processing methods. In this study, left ventricular and aortic intravascular phonocardiograms were obtained from six patients by catheter-mounted piezoelectric transducers. Phonocardiograms (PCGs) were bandpass filtered (50-500 Hz), digitized at 4kHz and analyzed by three separate techniques: (1) Short-Time Fourier Transform (STFT), (2) Fast Wavelet Transform (FWT), and (3) a pulse-counting time domain method based on an Order Statistic (OS) filter. The resulting time frequency distributions were employed to examine intra-patient and inter-patient acoustic variability. Results suggest that STFT and FWT provide comparable temporal and frequency resolution of cardiac acoustical events. However, the time-varying and multicomponent nature of heart sounds was poorly characterized by the OS technique employed in this study. This study suggests that improved localization of acoustic events during the cardiac cycle may prove useful in the development of automated auscultation devices.