Automated Diagnosis of Tachycardia Beats

Abstract

Due to tachycardia, heart generates lethal arrhythmia beats namely atrial flutter (AFL), atrial fibrillation (A-Fib), and ventricular fibrillation (V-Fib). These irregular patterns are very effectively and noninvasively reflected using standard electrocardiogram (ECG). In this study, an automated diagnosis support system (DSS) is developed for accurate discrimination and classification of complete classes of tachycardia beats (atrial as well as ventricular) using higher-order spectra (HOS). In this multiclass diagnosis problem, dimensionality of HOS third-order cumulants is reduced using independent component analysis (ICA) and fed for standard hypothesis test ANOVA (p < 0.05). Finally, statistical significant components are subjected for ensemble classification using random forest (RAF) and rotation forest (ROF) classifiers and to realize best performance tenfold classification is performed. Further, the consistency of classifiers is assessed using Cohen’s kappa matric. Proposed DSS achieved overall classification accuracy of 99.54% using ROF. Our reported results are highest than published in the earlier works.