A robust wavelet-based approach for dominant frequency analysis of atrial fibrillation in body surface signals

Abstract

Objective: Atrial dominant frequency (DF) maps undergoing atrial fibrillation (AF) presented good spatial correlation with those obtained with the non-invasive body surface potential mapping (BSPM). In this study, a robust BSPM-DF calculation method based on wavelet analysis is proposed.Approach: Continuous wavelet transform along 40 scales in the pseudo-frequency range of 3-30 Hz is performed in each BSPM signal using a Gaussian mother wavelet. DFs are estimated from the intervals between the peaks, representing the activation times, in the maximum energy scale. The results are compared with the traditionally widely applied Welch periodogram and the robustness was tested on different protocols: increasing levels of white Gaussian noise, artificial DF harmonics presence and reduction of number of leads. 11 AF simulations and 12 AF patients are considered in the analysis. For each patient, intracardiac electrograms were acquired in 15 locations from both atria. The accuracy of both methods was assessed by calculating the absolute errors of the HDF B SP M with respect to the atrial HDF, either simulated or intracardially measured, and assumed correct if ≤ 1 Hz. The spatial distribution of the errors between torso DFs and atrial HDFs were compared with atria driving mechanisms location. Torso HDF regions, defined as portions of the maps with |DF −HDF B SP M | ≤ 0.5 Hz were identified and the percentage of the torso occupied these regions was compared between methods. Main results:The proposed method allowed a significant improvement on non-invasive estimation of the atria HDF (median relative error of 7.14% vs. 60.00%, p = 0.06), outperforming the Welch approach in correct estimations of atrial HDFs non-invasively for both cases: models (81.82% vs 45.45%) and patients (75.00% vs 66.67%). A low positive BSPM-DF maps correlation was seen between techniques (0.47 for models and 0.63 for patients), highlighting overall differences in DF distributions. The method was more robust to white Gaussian noise and harmonics and presented more consistent results in lead layouts with low spatial resolution (p = 0.99 vs. p = 0.94).Significance: Estimation of atrial HDFs using BSPM is improved by the proposed waveletbased algorithm, helping increase the non-invasive diagnostic ability in AF.

Electrocardiographic Imaging for Atrial Fibrillation: A Perspective From Computer Models and Animal Experiments to Clinical Value

“…The same approach was applied by Guillem et al (2009a), specifically for recordings during AF with a very similar conclusion, with 34 leads as the limit for the signals to incorporate independent information above the noise level. Later, Marques et al (2020b) showed that it is possible to even characterize atrial rotors on the body surface with as few as 32 leads. A more recent study quantified the number of leads required for accurate localization of the origin of atrial or ectopic beats and concluded that 74 leads are necessary (Parreira et al, 2020).…”

“…Nevertheless, the combination of re-entrant analysis and highest dominant frequency regions has been suggested as an optimal approach to identify AF drivers from ECGI in a computational study by Rodrigo and colleagues (Rodrigo et al, 2017a). Non-invasive estimation of frequency of activation during AF showed to be more accurate with a wavelet-based approach rather than with Welch method in both models and AF patients (Marques et al, 2020b).…”

Classification and Location of Atrial Arrhythmic Mechanisms With Body Surface Potential Mapping

“…DFs on BSPM signals were estimated with a method based on continuous wavelet transform (CWT) [10]. Peaks detected in the outcomes of a CWT performed with a negative first-order Gaussian wavelet match patterns similar to singularities [10], such as the sharp positive deflections associated with depolarization wavefronts in BSPM signals [11]. CWT was applied for each lead along 40 linearly spaced scales in the pseudo-frequency range of 3 to 30 Hz ( Figure 1A).…”

Removing Clinical Motion Artifacts During Ventilation Monitoring With Electrical Impedance Tomography: Introduction of Methodology and Validation With Simulation and Patient Data

“…The DWT processing is shown in Figure 2. It is composed of two parts: decomposition and rebuild (12,14,15). In decomposition processing, using the so-called wavelet functions and the scaling functions, DWT decomposes the noisy EIT signal (corrupted by motion artifacts) into a relatively slowvarying signal a 1 (n) (approximation coefficients) and a fastvarying signal (detail coefficients) at the first step.…”