Abstract

Beats classification is an essential step in the ECG signal analysis for cardiac arrhythmias detection. There are multiple alternatives to solve this problem, but these are considerably reduced when re-al-time restrictions are added to the analysis. The goal of this work is to expose an optimal solution based mainly on the use of voltage values of the signal in the time domain and compare it with other based on Daubechies’ Wavelets analysis. Several measures are used in both feature spaces to determine the similarity of every beat to a patient’s specific patterns and, after that, a method similar to clustering’s algorithms is used to assign a class to each. To evaluate the performance of the pro-posed algorithms, ECG signal records extracted from the MIT-BIH data-base are used. With the method used in the analysis, we obtained 93.25% of sensitivity and 91.43% on premature ventricular contractions predictivity, which allow us to conclude that it is very feasible for their application in real time systems, due to their low computational cost.

Keywords: ECG; Beat classification; Real time; Cluster