Neural-Network Detection of Abnormal Ventricular Beats Using Temporal and Morphological Features of the Electrocardiographic Signal

摘要

In this paper we present a neural-network-based detecting system to identify premature ventricular contractions (PVC). We used two feed forward multilayer perceptron topologies (MLP): one with 6 nodes in the input layer to process only timing features of the ECG and another model with 8 input nodes to include time and shape features of the electrocardiographic signal. We trained and tested our neural network models using records from the MIT/BIH arrhythmia database. By adding morphological features as inputs to the network, its overall performance increases when compared to temporal features alone.