Abstract
This paper analyses a strategy for ischemia detection-based on wavelet decomposition of the ST segment. The wavelet transform is used as a pre-processing tool for linear discriminant classifier. In order to minimize generalization problems caused by correlations between the classification variables, a selection algorithm is employed to choose a subset of wavelet coefficients with appropriate discriminability and small collinearity. When applied to a set with small morphologic variability, good results are obtained: 98.5% of accuracy and a ROC area equal to 0.98 . However, when the training set has a high within-class scatter, the discriminant model yields poor results