Comparison of Artificial Neural Networks for Cardiac Arrhythmia Classification
Keywords:
Electrocardiogram (ECG), Arrhythmia, Denoising, Continuous Wavelet Transform (CWT), Artificial Neural Networks.Abstract
Arrhythmia is an anomalous electrical activity of the heart, and an electrocardiogram (ECG) is a method to
identify abnormalities in heart like arrhythmias. Due to some factors like presence of noise and nonstationary nature, it
is difficult to analyze and interpret an ECG signal. Usually computer-aided analysis of ECG results assists medical
experts to detect arrhythmias. In this work, ECG signal is denoised and features like RR-interval, average RR-interval,
pre- and post RR-intervals, R-amplitude, CWT coefficients after PCA reduction are used to classify arrhythmias. Here
six types of beat classes of arrhythmia as recommended by the Association for Advancement of Medical Instrumentation
are analyzed for classification, they are Normal beat (N), Left bundle branch block beat (LBBB), Right bundle branch
block beat (RBBB), Premature ventricular contraction (V), Atrial Premature beat (A) and Paced beat (P). Beats are
classified using artificial neural networks. Five different networks are used for classification, they are Feed forward
backpropagation, Elman backpropagation, Cascade forward backpropagation, Layer recurrent and NARX. The results
showed that Feed forward back propagation classified the N, L, R, V, A and P arrhythmia classes with high accuracy of
94.2%.
