Robust sequential detection algorithm for cardiac arrhythmia classification

We describe a modified sequential probability ratio test (SPRT) for the discrimination of ventricular fibrillation (VF) from ventricular tachycardia (VT) in measured surface electrocardiograms. The algorithm uses a novel regularity measure dubbed blanking variability (BV) applied to threshold crossings from the measured ECG. Blanking variability corresponds to the normalized rate of change of cardiac rate as the blanking interval is varied. The algorithm has been trained and tested using separate subsets drawn from the MIT-BIH malignant arrhythmia database. BV values are modeled using a truncated Gaussian distribution, and parameter values are derived by averaging over the training component of the database. In testing, the algorithm achieved an overall classification accuracy of 95%.