Common spatial pattern: An improved method for atrial fibrillation wave extraction

In this work, a spatial filtering by common spatial pattern (CSP) is used for atrial fibrillation extraction. The CSP technique was applied by defining intervals in the ECG time series containing either the QRST signal or the remaining part of the ECG. This enabled a decomposition of the 12 lead ECG into 12 components sorted according to their similarity to the QRST signal. A test database with simulated AF was generated. This test set was used to study the performance of several classical techniques and of the new method. The techniques considered were: CSP, template subtraction, PCA and the ICA variants: SOBI, JADE, INFOMAX. For PCA, ICA and CSP, the subset of components that best matched the original AF signal was considered for each method. After testing the methods proposed, the one that performed best with our dataset was the CSP. The method that gave the worst results was the template subtraction.