Multistage Classification of Current Density Distribution Maps of Various Heart States Based on Correlation Analysis and k-NN Algorithm

Abstract

Magnetocardiography is a modern method of registration of the magnetic component of electromagnetic field, generated by heart activity. Magnetocardiography results are a useful source for the diagnosis of various heart diseases and states, but their usage is still undervalued in the cardiology community. In this study, a two-stage classification by correlation analysis using a k-Nearest Neighbor (k-NN) algorithm is applied for the binary classification of myocardium current density distribution maps (CDDMs). Fourteen groups of CDDMs from patients with different heart states, healthy volunteers, sportsmen, patients with negative T-peak, patients with myocardial damage, male and female patients with microvascular disease, patients with ischemic heart disease, and patients with left ventricular hypertrophy, divided into five and three different groups depending on the degree of pathology, were compared. Selection of best metric, used in classifier and number of neighbors, was performed to define the classifier with best performance for each pair of heart states. Accuracy, specificity, sensitivity, and precision values dependent on the number of neighbors are obtained for each class. The proposed method allows to obtain a value of average accuracy equal to 96%, 70% sensitivity, 98% specificity, and 70% precision.