Spectral analysis of closing sounds produced by lonescu-Shiley bioprosthetic aortic heart valves

Abstract

The objective of the paper is to compare the performance of conventional FFT-based and modern parametric methods when extracting, from aortic closing sounds produced by lonescu-Shiley bioprosthetic heart valves, three features used in diagnosing valve dysfunction. Eight algorithms were tested by adding random noise and truncating 15 simulated aortic closing sounds. The performance of each algorithm was evaluated by computing the absolute error between the parameters obtained from the reference spectra of the simulated sounds and those obtained from the estimated spectra. Results show that the fast Fourier transform with rectangular window (FFTR) can locate the dominant spectral peak of the valve sound with an average accuracy of 10 Hz. Pole-zero modelling using the Steiglitz-McBride method with maximum entropy (SMME) is the best technique for estimating the frequency of the second dominant spectral peak and the bandwidth at −30 dB of the spectrum, with an average accuracy of 50 Hz and 27 Hz, respectively. In addition to this analysis, the accuracy of the frequency distribution of the estimated spectra was evaluated. Results show that the Steiglitz-mcBride method with extrapolation to zero and FFTR are the best algorithms to estimate the distribution of the reference spectra in the 20–200 Hz frequency bands. In the 200–500 Hz and 500–1000 Hz frequency bands, SMME gives the best results.