No abstract

The paper describes the design, training and testing of a three-layer feed-forward back-propagation neural network for the classification of bioprosthetic valve closure sounds. Forty-seven patients with a porcine bioprosthetic valve inserted in the aortic position were involved in the study. Twenty-four of them had a normal bioprosthetic valve, and the other 23 had a degenerated valve. Five features extracted from the Fourier spectra and 12 linear predictive coding (LPC) coefficients of the sounds were used separately as the input of two neural-network classifiers. The performance of the classifiers was tested using the leave-one-out method. Results show that correct classifications were 85 per cent using the spectral features, and 89 per cent using the LPC coefficients. The study confirms the potential of artificial networks for the classification of bioprosthetic valve closure sounds. Clinical use of this method, however, still requires further investigation.

A pattern recognition method using diagnostic features from the low frequency spectra of Bj6rk-Shiley convex0 concave (BSCC) valve opening sounds was tested for differentiating intact valves from those having a single leg separation (SLS) of the outlet strut. The method was applied to valve opening sounds obtained from a pulse duplicator system (6 intact and 6 SLS valves) and recorded in sheep (1 1 intact and 16 SLS valves). A total of 30 opening sou& were analyzed for each sheep and 10 opening sounds for each pulse duplicator test. For each valve, 21 diagnostic features were extracted from the mean power spectrum of the opening sounds for training and testing the K-nearest neighbor and the Bayes classifiers. All combinations of I, 2, 3 and 4 features among 21 were evaluated to find the most discriminant feature sets. For the in vitro study, 100% of correct classlfications (CCs) was obtained with severalfeature sets and either classrfier. For the sheep study, 93% of CCs was obtained with two feature sets and the Bayes classiper. The feature set that could best classifL both in-vivo and in-vitro valve opening sound spectra provided 82 % of CCs. The in vitro and in vivo studies thus demonstrated that the low frequency analysis of BSCC opening sounds is a promising method of detecting SLS of BSCC mitral valves. 1: IntroductionStein et al. in 1987 were the first to evaluate the potential of using spectral analysis of opening sounds produced by Bjork-Shiley convexo-concave (BSCC) disc valves as a noninvasive indicator of single leg separation (SLS) of the outlet strut [I]. An intact 29 mm BSCC valve and a normally functioning 29 mm BSCC valve with a SLS in the mitral position of a pulse duplicator were examined by low-frequency spectral analysis. Clear differences were observed between the opening sound spectra of the two valves. The dominant frequency of the opening sound of the intact valve was 234 Hz while that of the SLS valve was 78 Hz. The correlation between the frequency spectra of the opening sounds of the two valves was poor (correlation coefficient r = 0.22) indicating a marked difference in the spectral signature. In addition, comparison of the spectra of both valves showed that the maximal frequency of the SLS valve was lower than that of the normal valve. This in vitro observation suggested that frequency analysis of the opening sound of BSCC valves had potential for the noninvasive identification of SLS valves. The objective of the present study was to validate this hypothesis on a larger number of BSCC valves recorded in vitro and in vivo. 1063-7125/94 $3.00 0 1994 IEEE 61

No abstract