Optimal beat selection study for QSRT cancellation methods in the ECG of atrial fibrillation

The proper analysis and characterization of atrial fibrillation (AF) from surface ECG recordings requires cancelling out ventricular activity (VA), i. e. the QRST complexes. For single-lead ECGs, averaged beat subtraction (ABS) technique is the most widely used. However, this method is very sensitive to QRST wave variations. To date, several strategies have been proposed to overcome this problem, but an exhaustive study about the most suitable set of beats to obtain the most accurate QRST cancellation template, given a concrete recording, has not been performed yet. Thereby, a thorough study about the VA cancellation accuracy, when different complexes are selected to obtain the ventricular template, is presented in this work. The results showed that the most correlation-similar complexes to the one under cancellation provided a more accurate ventricular cancellation template than the preceding and subsequent beats, thus providing a higher AA extraction quality. In addition, a high number of complexes (greater than 30) is not recommended, because the AA extraction quality improvement is reduced. Nevertheless a more accurate VA representation is obtained when the number of complexes used to generate the QRST cancellation template is increased, but remains below 30.