APPLICATION OF RECURRENCE ANALYSIS FOR DETECTION OF MYOCARDIAL INFARCTION ON 12 LEAD ECG SIGNALS
ECG; KNN; Myocardial Infarction; Phase Space Reconstruction, Recurrence Quantification Analysis.
The use of nonlinear dynamic systems analysis methods to identify cardiac arrhythmias from electrocardiogram (ECG) signals has become widely disseminated, especially the technique of recurrence quantification analysis (RQA) of reconstructed phase space (PS). However, the choice of PS reconstruction parameters, delay and dimension, is still controversial in the literature. The main objective of this study is to evaluate the impact of the choice of delay and dimension on the classification of ECG signals in healthy patients and patients with myocardial infarction (MI). The classification was executed by producing a k-nearest neighbors (KNN) model trained from RQA attributes taken from two forms of PS reconstruction, one with optimal embedding parameters calculated for each signal and another with fixed embedding parameters. Complementally, the performances of the algorithm to detect MI in different affected cardiac regions were also evaluated, besides testing five different techniques of attribute normalization. The results indicate that despite the models with fixed embedding parameters, in general, having superior performance to the models with variable embedding parameters, rarely this difference was statistically significant. When the infarction location is taken into account, among the best results, the performance of the KNN model with fixed embedding parameters was superior in all cases where there was a statistically significant difference. In addition, we obtained a score of 0.815 on the ROC AUC for MI detection and the MinMax scaler was the most robust attribute normalization technique.