Arrhytmic and Non-Arrhytmic Heartbeat Classification based on Gaussian Mixture Model

Abstract

Probabilistic neural networks have been frequently used in classification of nonstationary and individual signal patterns due to its prediction capability rather than certain results. The purpose of this study is to perform the classification of heartbeat in arrhythmic and non-arrhythmic electrocardiogram (ECG) signal based on Gaussian mixture model and logarithmic linearization. To achieve this operation, a log-linearized Gaussian mixture network (LLGMN) was used and the training and comparison calculations attaining high performance were explained. The proposed method was applied on the EKG signals from MIT-DB database and also real subjects, and resulted the classification at the rates of 99.93% sensitivity, 96.32% positive predictive value and 0.71% error. The proposed algorithm is independent to the sampling rates of ECG device and individual frequency characteristics of the heart beats, and does not required prepossessing such as filtering. It also can be operated regardless of the type of arrhythmic or nonarrhythmic ECG signal. Based on the results, the proposed algorithm is capable of high performance classification for different types of ECG signals from various channels in various characteristics.