Bias and variability of diagnostic spectral parameters extracted from closing sounds produced by bioprosthetic valves implanted in the mitral position.

Abstract

A method is proposed to estimate the bias and variability of eight diagnostic spectral parameters extracted from mitral closing sounds produced by bioprosthetic heart valves. These spectral parameters are: the frequency of the dominant (F1) and second dominant (F2) spectral peaks, the highest frequency of the spectrum found at -3 dB (F-3), -10 dB (F-10) and -20 dB (F-20) below the highest peak, the relative integrated area above -20 dB of the dominant peak (RIA20), the bandwidth at -3 dB of the dominant spectral peak (BW3), and the ratio of F1 divided by BW3 (Q1). The bias and variability of four spectral techniques were obtained by comparing parameters extracted from each technique with the parameters of a spectral "standard." This "standard" consisted of 19 normal mitral sound spectra computed analytically by evaluating the Z transform of a sum of decaying sinusoids on the unit circle. Truncation of the synthesized mitral signals and addition of random noise were used to simulate the physiological characteristics of the closing sounds. Results show that the fast Fourier transform method with rectangular window provides the best estimates of F1 and Q1, that the Steiglitz-McBride method with maximum entropy (pole-zero modeling with four poles and four zeros) can best evaluate F2, F-20, RIA20 and BW3, and that the all-pole modeling with covariance method (16 poles) is best suited to compute F-3. It was also shown that both the all-pole modeling and the Steiglitz-McBride methods can be used to estimate F-10. It is concluded that a single algorithm would not provide the best estimates of all spectral parameters.