Application of the Wavelet Analysis of the Electrocardiograms for diagnostics of Paroxysmal Atrial Fibrillation

A new algorithm for completely automated wavelet transform (WT) analysis of RR-intervals featuring paroxysmal atrial fibrillation (PAF) has been developed. The power spectrum of wavelet-coefficients of RR-intervals sequence was intended to predict the onset of PAF. The absence of longterm wavelet-coefficients - correlation indicates on independence of RR-intervals from distant intervals and chaotic character of atrium excitation for the patients with PAF, in contrast to correlation of all RR-intervals for healthy patients. The predictor uses the diagnostic criterion: 5-10 time increase of the amplitude of wavelet-coefficients precedes the onset of paroxysm. WT based algorithm also was intended to differentiate ECG signals with non-terminating permanent AF and signals with paroxysmal AF (spontaneous self-terminated AF). Discrete WT of ECG has been used for identification of QRS-complexes on the basis of threshold criterion for corresponding wavelet coefficients for the purpose to remove the ventricular signal parts. Then we have decomposed with continuous WT the remainder ECG signal (the atrial activity) into different scales (frequency bands). The distribution of wavelet power spectrum is a discriminant diagnostic criterion of AF forms - a permanent or paroxysmal AF. If the WT power spectrum has a local maximum at scale higher than some predefine threshold parameter, AF is likely terminated within up to two minute, i.e. we can define paroxysmal AF. For ECG with WT power spectrum local maximum at scale lower than threshold parameter, AF is likely permanent AF.