Session P86.5

Comparative Study of Non-Invasive Organization Estimation Strategies to Predict Spontaneous Termination of Atrial Fibrillation

R Alcaraz*, JJ Rieta

Universidad Politecnica de Valencia
Valencia, Spain

To date, several methods to predict spontaneous paroxysmal atrial fibrillation (AF) termination from surface ECG recordings have been proposed. Three of them are based on the Sample Entropy (SampEn) non-invasive organization estimation of AF. However, each one of them uses different signal transformations to generate the final time series over which SampEn is applied. The present work compares these strategies making use of the same patient’s database.
In the first strategy, the atrial activity (AA) is obtained through QRST cancellation. Next, the main atrial wave (MAW) is obtained by selective filtering centered on the atrial dominant frequency, thus yielding the time series for SampEn computation. In the second strategy, an equivalent wave to the MAW is obtained by applying seven levels of discrete wavelet decomposition to the AA. The corresponding sub-band is reconstructed back to time domain and evaluated with SampEn. In the last strategy, the time series is obtained as the concatenation of TQ segments, free of QRST complexes.
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 labeled recordings of each group formed the learning set. The methods’ performance was evaluated by computing sensitivity and specificity with the ROC curve. For the learning set, sensitivity values were 100%, 80%, and 80% and specificity values were 90%, 90%, 100% for the methods based on selective filtering, wavelet transform and concatenation of TQ segments, respectively. Regarding the test signals, a sensitivity of 93.75% and a specificity of 85.71% were provided for the three methods. Additionally, the same recordings were incorrectly classified by the three strategies. These coherent outcomes allowed us to conclude that the three techniques can estimate robustly AF organization and predict successfully paroxysmal AF termination. Nonetheless, the best predictive ability was provided by the method based on selective filtering. Therefore, it could be considered that the MAW contains important information about spontaneous paroxysmal AF termination.

(Abstract Control Number: 34)