Learning ECG Patterns with the Aid of Multilayer Perceptrons and Classification Trees

This paper presents an approach based on the combination of multilayer perceptrons (MLP) and classification tree (CT) to recognising four electrocardiograms (ECG) patterns: normal, left bundle branch block (LBBB), right bundle branch block (RBBB) and premature ventricular contraction (PVC). This study utilises MIT/BIH arrhythmia database as training and testing data. We first apply MLP and CT respectively to recognise ECG patterns. Next, we collect the ECG signal features which are selected in splitting CT´s node, and feed these selected features into MLP for ECG pattern recognition. The aim is twofold: reducing the input attributes of MLP so as to lower computation burden, and understanding which heartbeat features play important roles in recognizing above four ECG patterns. To compare the effectiveness of proposed method, we considered the principal component analysis (PCA) that was frequently used to cut down the input dimension for pattern recognition. Comprehensive computer simulations will justify the feasibility of the proposed approach.