T-Wave End Pattern Classification Based on Gaussian Mixture Model

Abstract

Nowadays, probabilistic neural networks have been used to pattern discrimination in non-stationary biological signals with individual characteristics. The main objective of this study was to develop a neural network based on Gaussian mixture model and logarithmic linearization to classify the T-wave ends, which are of the major parts of the ECG signals, For this purpose, a comparison algorithm evaluating time-series signals was established, and the limitations of the high performance classification process was determined. The proposed algorithm has been tested on the data from 4 normal subjects and 22 additional normal data sets from MIT-DB database. After the improvement by the proposed algorithm, we observed that the T-wave ends were detected with 7.20 and 5.10 milliseconds of the mean values and 9.32 and 12.44 milliseconds of standard errors, when the data from real subjects and MIT-DB database, respectively. The results suggested that the proposed algorithm achieved a classification and discrimination of various ECG signals at a high performance level.