To improve the efficiency when checking the morphology of the large dataset of a dynamic ECG and cluster QRS complexes

a method for QRS clustering is realized using Hermite functions and self-organizing neural network with premature ventricular contraction in high proportion. Using MIT-BIH arrhythmia database

a QRS complex is decomposed into a QRS vector by Hermit functions. All the vectors are input into self-organizing neural network for clustering. Elements of the vectors are used to analyze the clustering result. Positive rate is used to give a statistics result. Method achieves an average true positive rate of 91. 2 % and an average false positive rate of 1. 03 %. The method is effective in clustering using self-organizing neural network. The aim of clustering normal beats and premature ventricular contraction beats is attained.