Selection of optimal parameters for ECG diagnostic classification

The authors investigated the problem of selecting parameters for inclusion in neural networks for diagnostic classification of the Frank lead ECG as normal or one of six disease conditions. A database of 486 100% accurate classified cases was randomly divided into a training set (67%) and a test set (33%). Using a total of 274 parameters as well as the age and sex data the authors determined the discriminating power of each parameter with receiver operator characteristic analysis as well as the rank correlation of all possible parameter pairs. On the basis of the discriminating power and rank correlation, a number of different parameter selection schemes were considered. The authors achieved best classification rates on the test data set by selecting parameters which were maximally discriminating and non correlated.