Wavelet blind separation: a new methodology for the analysis of atrial fibrillation from Holter recordings

• Main Authors: Sanchez, C., Rieta, J.J., Castells, F., Alcaraz, R., Millet, J.
• Format: Conference Proceeding
• Language: English
• Subjects: Atrial fibrillation; Blind source separation; Cardiology; Data mining; Discrete wavelet transforms; Signal processing; Time of arrival estimation; Wavelet analysis; Wavelet packets; Wavelet transforms
• DOI: 10.1109/CIC.2004.1442962

This study shows the possibility of atrial activity (AA) extraction from atrial fibrillation (AF) episodes in Holter registers using only two leads (V1 and V5) with a new technique, the Wavelet Blind Separation (WBS). The WBS increases the observed mixtures of the original signal from the decomposition of each lead into six transformed signals using a wavelet transform. These mixtures provide enough useful information to a Blind Source Separation implemented system and the extraction can be achieved. To evaluate the suggested algorithm, artificial AF signals have been synthesized adding fibrillation activity to normal sinus rhythm. Results indicate that the WBS technique can be an important tool to perform the atrial extraction in short duration registers with a reduced number of leads like paroxysmal atrial fibrillation, which have to be usually detected from Holter systems.