Evaluation of Karhunen-Loeve expansion for feature selection in computer-assisted classification of bioprosthetic heart-valve status

This paper analyses the performance of four different feature-selection app roaches of the Karhunen-Loeve expansion (KLE) method to select the most dis criminant set of features for computer-assisted classification of bioprosth etic heart-valve status. First, an evaluation test reducing the number of i nitial features while maintaining the performance of the original classifie r is developed. Secondly, the effectiveness of the classification in a simu lated practical situation where a new sample has to be classified is estima ted with a validation test. Results from both tests applied to a reference database show that the most efficient feature selection and classification (greater than or equal to 97% of correct classifications (CCs)) are perform ed by the Kittler and Young approach. For the clinical databases, this appr oach provides poor classification results for simulated 'new samples' (betw een 50 and 69% of CCs). For both the evaluation and the validation tests, o nly the Heydorn and Tou approach provides classification results comparable with those of the original classifier (a difference always less than or eq ual to 7%). However, the degree of feature reduction is particularly variab le. The study demonstrates that the KLE feature-selection approaches are hi ghly population-dependent. It also shows that the validation method propose d is advantageous in clinical applications where the data collection is dif ficult to perform.