Diagnostic model for identification of myocardial infarction from electrocardiography signals

Abstract

Electrocardiography (ECG) is a useful test used commonly to observe the electrical activity of a heart. Recently, a growing relationship has been observed between diagnosis of any heart disease and using of machine learning techniques. In this scope, a diagnostic application model designed based on a combination of Recursive Feature Eliminator (RFE) and two different machine learning algorithms called 𝑘-nearest neighbors (𝑘-NN) and artificial neural network (ANN) is proposed for classification of ECG signals in this study. The experiments performed on an open-access ECG database. Firstly, the signals were passed a pre-processing step. Then, several diagnostic features from morphological and statistical domains were extracted from ECG signals. In the last stage of the analysis, RFE algorithm covering 10- fold cross-validation and the mentioned machine learning techniques were employed to separate Myocardial Infarction (MI) samples from normal. The promising results as accuracy of 80.60%, sensitivity of 86.58% and specificity of 64.71% were achieved. The validation of the contribution was checked by comparing the performances of both 𝑘-NN and ANN to related works. Consequently, the proposed diagnostic model ensured an automatic and robust ECG signal classification model.