Catheter Ablation -> Atrial Fibrillation & Atrial Flutter: -> Mapping & Imaging D-PO04 - Poster Session IV (ID 15) Poster

D-PO04-182 - Extracting Atrial Electrograms From Unipolar Mapping Signals In Atrial Fibrillation (ID 485)

 B.H. Smaill: Nothing relevant to disclose.


Background: Atrial unipolar records are contaminated by drift, mains noise and ventricular components. Construction of bipolar signals as differences between adjacent electrodes reduces but may not fully remove these artifacts.
Objective: The objective of this exercise was to design a comprehensive wavelet-based approach for recovering unipolar atrial electrograms (EGMs).
Methods: Simultaneous EGMs from 14 patients with persistent AF acquired with 64-channel Constellation® catheters in the LA and a coronary sinus (CS) catheter were processed off-line in 3 stages following mains noise removal. 1) Ventricular activation was identified from the prominent R waves in the CS signal. A continuous wavelet transform (CWT) for each LA signal was computed across time windows -50 to +400 ms relative to R with a 2nd order Gaussian mother wavelet at 10 scales. Using differentially weighted CWT components, QRS and T waves were removed smoothly, but high frequency components were subtracted at the QRS peak only. 2) Atrial activation was identified from the CWT computed with a 1st order Gaussian mother wavelet at 10 scales. Power spectra estimated at each wavelet scale were combined to provide a robust estimate of instantaneous signal power relative to an adaptive threshold, with maxima identifying activations. 3) EGMs could then be reconstructed by combining weighted wavelet components across specified time windows defined by the activations detected.
Results: Wavelet-based QRS-T subtraction is fast and robust. Subtraction residue is low with RMS error 1.6 +1.1% (median +interquartile range) significantly better than than fixed template subtraction of time-window interpolation methods (p<0.05). Atrial activation detection replicates the judgement of expert observers and enables a wide range of EGM morphologies to be extracted and tagged when signal-to-noise ratios are low.
Conclusion: Wavelet-based filtering is a powerful way of extracting timing information and EGM morphology from noisy unipolar signals in AF. Separate ventricular components in each channel are detected and removed, and atrial activation is unmasked via wavelet functions that replicate key EGM morphologies and differential weighting of CWT components across time scales.