Algoritmo de identificação automática dos batimentos cardíacos e reconhecimento das contrações ventriculares prematuras em ECG utilizando inteligência artificial

Abstract

The aim of this paper is to develop an algorithm that can identify heart beats based on Pan & Tompkins’ work. The identified heart beats were used as a reference to extract the features for the KNN classifier used to recognize the premature ventricular contraction. It was changed in the Pan & Tompkins’ work, the band pass and derivative filters. A new stage was included that decides which peak in a range of 100 units is a R peak by analysing its amplitudes. The accuracy of the R peak algorithm was 99.66 %, sensibility of 99.78 % and positive prediction of 99.68 %. It was extracted from the hearts beat 46 features. Using genetic algorithm, this amount was reduced to 26. The accuracy of the algorithm 1NN was initially 80.70 % using all features. With 26 features chosen by the genetic algorithm, this accuracy was enhanced to 97.87%. To be possible to use the genetic algorithm, it was used only 36 % of the all data. After the feature selection, the R peaks identified was classified as PVC and not PVC. The accuracy of the KNN algorithm was 98.10%, sensibility of 99.20% and positive prediction of 98.85%.