Time-frequency analysis and classification of heart sounds and murmurs

Abstract

Heart sounds and murmurs are time-varying and non-stationary signals. It can show the difference between normal heart and pathological heart murmur. The signal gathered from heart auscultation and phonocardiogram do not provide permanent record of examination result for future evaluation. In addition, the auscultation needs a skilled physician to determine the heart condition correctly. Time-frequency distribution (TFD) is a method that can represent non-stationary and time-varying signals. Contrast to time-domain or frequency-domain, TFD is able to show the variation in the frequency content of the signal with time. In this study, the selected TFD for analysis purposes are the Wigner-Ville distribution (WVD), windowed Wigner-Ville dis$bution (WWVD) and smooth windowed Wigner-Ville distribution (SWWVD). The main contribution is to determine the distribution that accurately shows the time-frequency representation of heart sounds and murmurs. Comparison is made based on the mainlobe width (MLW), peak-to-sidelobe average ratio (PSAR) and signal-to-interference ratio (SIR). In general, the SWWVD shows the most accurate time-frequency representation based on the SIR which achieved 16.40 dB for normal heart compared to -4.82 dB using the WVD. From the time-frequency representation, a further operation of signal detection using the Moyal's formula can be applied. The Moyal's forrnula is used to classify the murmurs in the presence of noise. On the average, time-frequency classification performs better over the timedomain correlation by -9.94 dB of noise power added for all signal used. This is mainly due to the non-stationarity of the signal which is good to be analyzed in TFD.