Session P79.5

Robust Prediction of Atrial Fibrillation Termination Using Wavelet Bidomain Entropy Analysis

R Alcaraz*, JJ Rieta

Universidad de Castilla-La Mancha
Cuenca, Spain

In this work a robust method to predict atrial fibrillation (AF) termination has been developed by estimating, through regularity indexes, the atrial activity (AA) organization increase prior to AF termination. Regularity variation appears as a consequence of the decrease in the number of reentries wandering the atrial tissue before AF termination. AA was obtained from surface ECG recordings using a QRST cancellation technique. Next, Wavelet transform (WT) was applied in a bidomain way (time and scale) in order to improve regularity estimation performance and minimize noise. The AA organization was estimated by Sample Entropy (SampEn) both in time and wavelet domains (bidomain) from the scale containing the dominant atrial frequency. Finally, a robust and reliable classification process for spontaneous terminating and non-terminating AF episodes was developed making use of two different wavelet decompositions strategies and the dominant atrial frequency.
Fifty recordings available at Physionet were used. 26 of them were non-terminating AF episodes and 24 were AF episodes terminating immediately after the end of the recording. 10 labelled recordings of each group formed the learning set. By applying the methodology to the learning sets in the time domain, 80% sensitivity, 100% specificity, and 0.0902 as optimum SampEn threshold were obtained. On the other hand, 100% sensitivity, 70% specificity, and 0.034 as optimum SampEn threshold were obtained in the wavelet domain. Finally, by applying the combined bidomain methodology 28 out of 30 test signals were correctly classified. Therefore, the AF behavior of 48 out of 50 recordings (96%) was correctly predicted, and 96.15% sensitivity and 95.83% specificity were obtained. Results also indicated in time domain that terminating episodes presented lower SampEn values (0.0783±0.0098) and, therefore, higher organization than non-terminating episodes (0.0951±0.0123). Regarding the wavelet domain, terminating episodes presented a more irregular wavelet coefficients vector (0.0356±0.0053) than non-terminating episodes (0.0260±0.0063).

(Abstract Control Number: 51)