Fundamental frequency estimation in atrial fibrillation signals using correntropy and Fourier Organization Analysis

Advanced analysis of atrial fibrillation (AF) intra - cardiac electrograms aims to establish clinical targets for ablation. Frequency domain approaches, such as Dominant Frequency Analysis (DFA), estimate the dominant frequency to identify cardiac sites with high activation rates as ablation targets. However, they often discard relevant information in the spectrum, such as the harmonic structure or the spectral envelope. Moreover, these methods do not provide a complete characterization of complex atrial fibrillation signals. We propose to use the correntropy function to estimate the fundamental frequency, instead of dominant frequency, in order to quantify the activation rate of AF signals. We also propose to use the correntropy method combined with Fourier Organization Analysis (FOA), a method previously proposed to model fibril-latory signals, in order to estimate more than one fundamental frequency in complex signals which can be the result of the interaction of multiple wavefronts. The proposed approach was benchmarked by using pseudo-real AF signals for assessing their performance in a controlled environment. Two datasets of AF real signals were assembled, one with regular and simple signals, and another one with complex signals. Correntropy estimation of fundamental frequency, combined with FOA, allowed not only to characterize the periodicity of AF signals, but also to study more complex signals by modelling them with more than one component.

I. Introduction

ATRIAL fibrillation (AF) is nowadays the most common sustained arrhythmia. Catheter ablation is a useful method to prevent recurrences in patients with paroxysmal or persistent AF, and recently, analysis of intra–cardiac electrograms has been used as a criterion for selecting efficient ablation targets [1], [2]. Two main approaches have been used to guide the ablation procedure to date, namely, time domain [3] and frequency domain methods [4], the former based on activation rate, and the latter based on dominant frequency representing the inverse average cycle length. Both methods aim to estimate the electrogram periodicity properties with the objective of finding those regions with the shortest average cycle length. A common technique used to perform the frequency domain analysis of AF electrograms is the so-called Dominant Frequency Analysis (DFA) [5], [6], which aims to estimate the average excitation rate (inverse of mean cycle length) based on the dominant frequency of a preprocessed AF signal. However, such descriptions have often discarded relevant information of the spectrum, e.g., the harmonic structure or the spectral envelope. Additionally, it is not guaranteed that a dominant frequency will provide a good estimation of the averaged cycle length, given that there are several controversial studies about these issues [7], [8]. More, it has been pointed out that DFA may not give a correct description when dealing with complex fractionated AF electrograms [9], [10].