QRS morphological classification using artificial neural networks

Artificial neural networks (ANNs) were applied to electrocardiographic (ECG) signals to classify QRS complexes. Several ANN paradigms were considered, and two were selected for the ECG analysis: backpropagation (BP) and the Kohonen feature map (KFM). ANNs were trained on 8 groups of 20 QRS complexes each, extracted from the VALE database (DB); each group was related to a QRS morphology as obtained by the DB annotations. The ANN performances were evaluated using both the learning set and the whole case as a recall set. The BP network showed a good specificity and was found able to separate morphologies with ambiguous DB annotations. The KFM network was able to create a clustering of QRS morphologies with a high agreement with the original annotations