Robust Atrial Ectopic Beat Classification From Surface ECG Using Second-Order Blind Source Separation

Abstract

Ectopic beats (EBs) generated from the atria or pulmonary veins are an important trigger mechanism for atrial fibrillation (AF). They can be periodic, and have been commonly observed during AF episodes. Robust noninvasive detection of EBs could improve pre-operative prediction, as well as post-ablation management. By separating periodic sources from surface ECG using second-order blind source separation methods, EBs were extracted, and discriminated from AF reentries, another type of periodic source. Our method is robust to noise of up to 0.5mV in the ECG, achieving an area-under curve of receiver operating characteristic (AUC-ROC) of 0.89±0.01 over a synthetic dataset of 31 reentries and 58 EBs, with and without Acetylcholine modulation in the left atrium.