Automated assessment of atrial fibrillation

This study presents our entry in this year's "Computers in Cardiology" challenge. The challenge is the automated assessment of the electrocardiogram for predicting the onset of paroxysmal atrial fibrillation/flutter (PAF). By considering a large set of features derived from R-R intervals, the P-wave shape and frequency representations of the P-wave, we compared the performance of a linear discriminant classifier processing the feature sets. A cross-validation scheme was used to estimate the classifier performance. The results demonstrated that features based on R-R intervals were the most successful. Independent testing showed that the ECG may be of some potential use in screening subjects for PAF and predicting the onset of PAF.