Time Frequency Analysis of Canine Heart Sounds

Six different time frequency distributions (TFDs) have been used to analyze the fust and second heart sounds. These TFDs have been applied to both simulated heart sound data and to experimeutal phonocardiogram data (N = 6) recorded in the anesthetized dog. Using the distributions, no significant systematic rising frequency component was observed in either the frst or second heart sounds. As revealed using the CWT, the energy of the lower frequency components are found to decrease more rapidly than the higher frequency components as the myocardium is progressively depressed by applying increasing concentrations of the gaseous anesthetic agents, halothane and isoflurane. INTRODUCTION AND METHODS Traditional methods have analyzed the phonocardiogram (PCG) signal visually in the time domain and used conventional Fourier techniques to examine spectral properties of the signals. In a study of canine fust heart sounds recorded from the epicardium [l] used the Binomial reduced interference distribution to observe a strong rising frequency component in early systole. Other researchers did not observe this component. Indeed, in a more recent study by [2], recordings from 27 sites across the human thorax showed that the fust heart sound consisted only of quasistationary and impulse-like components. In order to gain a greater understanding of the genesis of the fEst and second heart sounds; we investigated transthoracic canine heart sounds recorded from the apex region using six different time-frequency distributions; Spectrogram, Minimum Mean Cross Entropy (MMCE) combination of Spectrograms, Continuous Wavelet Transform (CWT), Wigner-Ville Distribution (WVD), Choi-William Distribution (CWD) and Zhao-Atlas-Marks Distribution (ZAM).. These distributions were also tested on simulated beart sound data. The aims were to compare and identify optimal timefrequency distributions (TFDs) for the characterization of the heart sounds and from this characterization assess the likelihood of a rismg frequency component in the fust and second heart sounds. A secondary aim was to assess the effects of the anesthetic agents, halothane and isoflurane on the heart sounds and to correlate these effects with their cardiodepressant action as determined by measurement of cardiac output. To investigate these effects, moment-based time-frequency measnres including signal localization in time and frequency, Heisenberg-Gabor Principle and the Reuyi-entropy are also presented. RESULTS AND DISCUSSION Simulated fust and second heart sounds generated to test the above distributions, as well as experimeutal data, showed that ZAM and CWT are able to provide the hest representation. The two major frequency components of the fust heart sounds are resolved clearly. For timefrequency representation of the second heart sound, due to the important time information regarding the separation of closure of the aortic (A2) and pulmonary (PZ) valves, ZAM was found to be superior. However, in visualizing TFDs as an energy density function, the CWT is the best because it is manifestly positive and provides good resolution. The Spectrogram is limited by insufficient resolution because its time and frequency resolutions cannot be simultaneously optimized. Attempts to use the MMCE combination of Spectrograms to improve the TFD gave little improvement and the use of the WVD is limited by extensive presence of cross-terms. The CWD suppresses some of the cross-terms in the representation of the heart sounds and its representation is able to provide better time and frequency localization with little smearing. Using the six distributions, no significant systematic rising frequency component was observed in either the fust or second heart sounds. Analysis of experimental heart sounds witb the intervention of anesthetic agents revealed that the energy of the heart sounds decreased with the concentration of the anesthetic agents as would be expected due to their cardiodepressant effects. As revealed using the CWT, the energy of the lower frequency components are found to decrease more rapidly than the higher frequency components as the myocardium is progressively depressed. In addition, the time localization of the frequency Components is progressively delayed as the myocardium is depressed with the application of the anesthetic agents.