Skip to Main Content
To separate the atrial (AA) from the ventricular (VA) electrical activity in surface ECG recordings of atrial fibrillation (AF), various methods have been proposed, such as QRS-T cancellation by beat-averaged template subtraction, and blind source separation (BSS). Although QRS-T cancellation is computationally more efficient than BSS, and allows the preservation of spatial information, it is sensitive to morphology changes, which produce large residuals in AA, biasing the frequency analysis. Aim of this study was: (i) to propose an improved approach to VA cancellation based on k-means morphology clustering (MC); (ii) to validate its ability to estimate AF dominant frequency (DF) on a standard database with intra-cardiac and surface ECG recordings (IAFDB, Physionet.org); (iii) to compare the temporal evolution of the spectral content of MC-estimated AA (MC-AA) with the one obtained from a reference BSS method based on Independent component analysis (ICA) and second-order blind identification (SOBI), in 14 body surface potential map (BSPM) recordings. QRS-T amplitude in MC-AA was significantly lower (p<;0.001) than in ECG (in closest BSPM channel to V1). The validation on IAFDB showed no significant difference in DF estimation (p=0.546) in 17 recordings. Also no significant difference in DF estimation (p=0.208) with respect to the reference BSS method was observed. The proposed QRS-T cancellation method effectively suppresses VA and accurately estimates DF compared to an established BSS method.