Atrial fibrillation analysis based on blind source separation in 12-lead ECG data

Abstract

Atrial Fibrillation is the most common sustained arrhythmia encountered by clinicians. Because of the invisible waveform of atrial fibrillation in atrial activation for human, it is necessary to develop an automatic diagnosis system. 12-Lead ECG now is available in hospital and is appropriate for using Independent Component Analysis to estimate the AA period. In this research, we also adopt a second-order blind identification approach to transform the sources extracted by ICA to more precise signal and then we use frequency domain algorithm to do the classification. The strategy used in this research is according to prior knowledge and is different from the traditional classification approach which training samples are necessary for. In experiment, we gather a significant result of clinical data, the accuracy achieves 75.51%. © 2010 Springer-Verlag.