A Bayes model for automatic detection and quantification of bioprosthetic valve degeneration

Abstract

The performance of a pattern recognition algorithm using a Bayes model for detecting bioprosthetic valve degeneration was evaluated using diagnostic features extracted from the sound spectra of 57 normal and 47 degenerated porcine xenograft valves. The data base was subdivided into training sets and test sets composed of various combinations of diagnostic features. The training sets were used to select the feature subset that minimized the classification error. The performance of the Bayes' classifier was evaluated with the test set by computing the percentage of correct classifications, false positives and false negatives. The test set was then exchanged with the training set and the classifier re-trained and tested. Mean values of correct classifications, false negatives and false positives were then computed. Results show that a mean correct classification rate of 74% was obtained for the automatic classification of normal and degenerated bioprostheses. The percentage of false positives was 13% while that of false negatives was 40%.