Classification of atrial fibrillation episodes by means of phase variations of time-frequency transforms

This study aimed to assess an early classification of paroxysmal and persistent atrial fibrillation (AF) episodes by means of the surface ECG on a heterogeneous cohort of patients (in terms of antiarrhythmic treatment and state of evolution of the arrhythmia), which is similar to the context that clinicians find at tertiary centres in their daily work. 129 consecutive unselected patients suffering from an AF episode conformed the study population (23 paroxysmal and 106 persistent). Modulus and phase features extracted from several time-frequency transforms of the ECG were studied, and it was phase variations which arose as determinant providing the best classification results using a Linear Discriminant Analysis classifier trained with 20 signals. Obtained performances for the latter feature were: Accuracy = 83.5% (total correct classifications), Sensitivity = 78.6% (paroxysmal AF episodes correctly classified), Specificity = 84.2% (persistent subjects properly classified). This results would aid electrophysiologists to choose and prescribe the most suitable treatment to lower recurrence and stop the natural progression of the arrhythmia in general scenarios.