Identification of supraventricular and ventricular arrhythmias using a combination of three neural networks

The authors present a classifier that makes use of a combination of three artificial neural networks to identify supraventricular and ventricular arrhythmias from two-lead intracardiac electrograms (ICEGs). Timing features measured from the right ventricular apex (RVA) and the high right atrium (HRA) leads were used to classify 8 rhythms of ICEGs into 4 categories. The decomposition of the classification problem into three easier-to-manage subproblems is discussed. A shared multilayer perceptron (MLP) architecture is presented that leads to an economic hardware implementation. The authors compare the classification performance achieved using the decomposition approach with that of a single large MLP. Simulations on data from 51 patients shows that the decomposition approach can achieve 95.1% to 96.2% correct classification on a separate testing data set