Study on the optimal use of Generalized Hurst Exponents for noninvasive estimation of atrial fibrillation organization

The Generalized Hurst Exponent H(q) relates to the existence of long-term self-dependencies in the autocorrelation function of a time series. In the present work the optimal use of H(q = 2) in the study of Atrial Fibrillation (AF) organization from noninvasive ECG recordings has been determined. With this aim, 60 signals from the Physionet AF Termination Database were studied. First, the optimal data length for computing H(2) was determined. Next, the application of a previous band-pass filtering step was assessed in order to improve the metric's performance. This improvement is due to the reduction of noise and ventricular residua in the signal that could affect the performance of nonlinear metrics. Finally, the use of H(2) in short recordings was studied analyzing only the last seconds of each recording. The use of a previous band-pass filtering stage improved significantly the performance of the metric and a classification accuracy of 95% was reached. In addition, the same classification accuracy was obtained in the analysis of the last 15 seconds of each recording, showing that H(2) can be applied in the study of short ECG recordings.