QRST cancellation using bayesian estimation for the auricular fibrillation analysis

The accurate extraction of the atrial fibrillation (AF) signal from the ECG is a challenging task. The performance of its frequency analysis is based on this accuracy. When the atrioventricular node modelling is addressed, the precision of the AF amplitude estimation plays an important role. In this work, we propose an AF estimator based on a bayesian approach. The a priori knowledge necessary to achieve good estimation performances is shown to be weak. The presented method is compared to classical ones such as the Average Beat Substraction (ABS) but also to modern one that is the Blind Source Separation (BSS) technic. The performances of the methods are illustrated on a small set of examples. The conclusion is that the proposed method gives an accurate global estimation of the AF signal but also locally since our results do not exhibit spurious oscillations at the time location of the R wave.