Premature Ventricular Contraction (PVC) Classifications by Probabilistic Neural Network (PNN) Using the Optimal Mother Wavelets

摘要

This paper presented our attempt to determine the reliability and accuracy of classifying Premature Ventricular Contraction (PVC) and several other arrhythmias using optimal mother wavelets and feature dataset obtained from our previous study in [1J. The proposed classifier is Probabilistic Neural Network (PNN) with less-overlapping data set between training and testing. In our previous study of |1|, we found that the most outperformed wavelets among the 35 mother wavelets tested are "haar", "db3" and "sym3" with overall average accuracy percentage of 85.47% for "haar" and 84.13% both for "db3" and "sym3". The result is slightly lower (<90%) than expected, as we found that the statistical indices of the wavelet coefficients used might not be good features; instead, using the whole coefficients may give higher accuracy. However, the calculation of peak-to-peak ratio proves to be encouraging as it provides convenient differentiator and is believed to be one of the factors that contribute to high accuracy. Addition to that, the selection of inverted R peak for PVC that do not have R peak also plays important role. It is observed that the accuracy of PVC with no R peak (inverted P. peak detection) is to be 91 28% fur "l.aar", *2.I9% Tor "db3" and 92.19% for "sym3".