Statistical Analysis of RR Interval Irregularities for Detection of Atrial Fibrillation

We compare two Atrial Fibrillation (AFIB) detection methods from surface ECG, based on RR interval variability in a statistical framework. We obtain the histogram of normalized RR differences for AFIB and non-AFIB episodes using MIT-BIH Arrhythmia database. Two probability density functions (pdf) are employed to model the histograms: Gaussian and Laplace. We then use Neyman-Pearson (NP) detection approach to obtain criteria for AFIB detection. The performance of the two methods is compared using Receiver Operating Curves (ROC) over the different databases. The result shows that the Laplace pdf approximates the histogram of normalized RR differences better than the Gaussian pdf and leads to better AFIB detection performance.