Spectral analysis of phonocardiographic signals using advanced parametric methods

• Main Author: Sava, Herkole P.
• Published: University of Edinburgh 1995
• Subjects: 615.84
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.661599

The research detailed in this thesis investigates the performance of several advanced signal processing techniques when analysis heart sound, and investigates the feasibility of such a method for monitoring the condition of bioprosthetic heart valves. A data-acquisition system was designed which records and digitises heart sounds in a wide variety of cases ranging from sounds produced by native heart valves to mechanical prosthetic heart values. Heart sounds were recorded from more than 150 patients including subjects with normal and abnormal native, bioprosthetic, and mechanical prosthetic heart values. The acquired sounds were pre-processed in order to extract the signal of interest. Various spectral estimation techniques were investigated with a view to assessing the performance and suitability of these methods when analysing the first and second heart sounds. The performance of the following methods is analysed: the classical Fourier transform, autoregressive modelling based on two different approaches, autoregressive-moving average modelling, and Prony's spectral method. In general, it was also found that all parametric methods based on the singular value decomposition technique produce a more accurate spectral representation than the conventional methods (i.e. Fourier transform and autoregressive modelling) in terms of spectral resolution. Among these, the Prony's method is the best. In addition a modified forward-backward overdetermined Prony's algorithm is proposed for analysing heart sounds which produces an improvement of more than 10% over previous methods in terms of normalised mean-square error. Furthermore, a new method for estimating the model order is proposed for the case of heart sounds based on the distribution of the eigenvalues of the data matrix.