Robust arrhythmia classifier using hybrid multilayered perceptron network

The paper describes a robust approach to model cardiac arrhythmias using the hybrid multilayered perceptron (HMLP) network. Healthy, cardiomyopathy, as well as left and right bundle branch block electrocardiograms (ECG) was obtained from the PTB Diagnostic ECG database. The signals were initially pre-processed for noise removal and baseline correction. 24 morphological descriptors from the bipolar limb leads were used as input to the neural network. 400 beat samples were obtained for each condition. Results show that the Levenberg-Marquardt algorithm attains the fastest convergence. Varying the number of hidden nodes however, has no significant effect on the classification accuracy. Performance comparison shows that the HMLP network is more robust and gives better classification accuracy over the multilayered perceptron (MLP) network. The error convergence meanwhile, indicates a leveled performance.