A nonlinear transformation of the ECG constituent patterns has been developed. The transformed ECG is mapped on the Euclidean two-dimensional plane and then a classification algorithm based on neural networks is used to identify the constituent patterns of the ECG. This way we can detect morphology changes of waves and visualize the results of classification. This cartography method could be used to standardize classification algorithms with linear or non-linear separating methods. The identification of tachycardia, ischemia and other heart diseases is easily done checking the appropriate areas of the resulting maps. The algorithm permits the on-line training of the classification scheme, when ambiguity concerning the pattern characterization arises.
展开▼