Are nonlinear ventricular arrhythmia characteristics lost, as signal duration decreases?

Abstract

A novel, nonlinear, phase space based method to quickly and accurately identify life-threatening arrhythmias is proposed. The accuracy of the proposed method in identifying sinus rhythm (SR), monomorphic ventricular tachycardia (MVT), polymorphic VT (PVT), and ventricular fibrillation (VF) for signals of at least 0.5 s duration was determined for six different ECG signal lengths. The ECG recordings were transformed into a phase space, and statistical features of the resulting attractors were learned using artificial neural networks. Classification accuracies for SR, MVT, PVT and VF were 93-96, 95-100, 79-91, and 81-88%, respectively. As expected, classification accuracy for the proposed method was essentially equivalent for ECG signals longer than 1 s. Surprisingly, classification accuracy for this new method did not degrade for 0.5 s ECG signals, indicating that even such short duration signals contain structures predictive of rhythm type. The phase space method's classification accuracy was higher for all segment durations compared to two other methods.