Atrial fibrillation onset prediction using variability of ECG signals

This study presents the use of two different methods for the automatic prediction of the onset of paroxysmal atrial fibrillation (PAF) by means of surface electrocardiographic (ECG) signal. The first method is commonly used and consists in the analysis of the heart rate variability (HRV) of the ECG signal. Two significant parameters are taken into consideration: the time domain metric standard deviation of average five minute window of the time series (SDANN) and the frequency-domain metric low-frequency/high-frequency (LF/HF) ratio of the RR interval. The second analysis method, which is based on the morphological timing characteristics of the QRS complex, is called morphologic variability (MV) of the ECG signal, and was not used before for PAF prediction. Parameters similar to those of HRV analysis are determined in this case. Both methods are applied on 198 Holter records taken from the PAF Database from physionet.org portal. The results show a better accuracy of the MV analysis than that obtained by means of HRV technique alone. Moreover, by using an appropriate decision rule, both methods were “fused” and the overall accuracy of PAF onset prediction was raised up to 90%. Experimental results also indicate that our method is applicable for usual Holter recordings and is robust against noise and common artifacts. Its high prediction accuracy is comparable with that obtained by manual annotation made by experts and therefore is suitable to be used in clinical practice.