Atrial Activity Extraction for Atrial Fibrillation Analysis Using Mutual Information Minimization

In this paper, we develop a new non-invasive method for atrial activity estimation in atrial fibrillation (AF) analysis. Our central approach is base on minimization of mutual information. To establish an appreciate separation model for independent component analysis (ICA), statistical properties of bioelectric independent sources corresponding to the ventricular and atrial activity (AA) are analyzed. During evaluation with real AF data, the proposed method is validated using two spectral parameters extracted from AA signals: main peak frequency (MPF) and spectral concentration (SC). Our results show that MPF ranges from 3.63 to 7.56 Hz and SP has a mean of 42.73, which suggest that the proposed method for AA extraction is very efficient.