Development of a toolbox for electrocardiogram-based interpretation of atrial fibrillation.

BACKGROUND: Atrial fibrillation (AF) develops as a consequence of an underlying heart disease such as fibrosis, inflammation, hyperthyroidism, elevated intra-atrial pressures, and/or atrial dilatation. The arrhythmia is initiated by, or depends on, ectopic focal activity. Autonomic dysfunction may also play a role. However, in most patients, the actual cause of AF is difficult to establish, which hampers the selection of the optimal mode of treatment. This study aims to develop tools for assisting the physician's decision-making process.

METHODS: Signal analytical methods have been developed for optimizing the assessment of the complexity of AF in all of the standard 12-lead signals. The development involved an evaluation of methods for reducing the signal components stemming from the electric activity of the ventricles (QRST suppression). The methods were tested on simulated recordings, on clinical recordings on patients in AF, and on patients exhibiting atrial flutter (AFL) and atrial tachycardia. The results have been published previously. Subsequently, the implementation of the algorithms in a commercially available electrocardiogram (ECG) recorder, an implementation referred to as its AF-Toolbox, has been carried out. The performance of this implementation was tested against those observed during the development stage. In addition, an improved visualization of the specific ECG components was implemented. This was enabled by providing a separate view on ventricular and atrial activity, which resulted from the steps implied in the QRST suppression. Furthermore, a search was initiated for identifying meaningful features in the cleaned up atrial signals.

RESULTS: When testing the implementation of the previously developed methods in the Toolbox on simulated and clinical data, the suppression of ventricular activity in the ECG produced residuals down to the level of physiologic background noise, in agreement with those reported on previously. The QRST suppression resulted in a better visualization of the atrial signals in AF, atrial AFL, sinus rhythm in the presence of atrioventricular blocks, or ectopic beats. Classifiers for AF and AFL that have been defined so far include the distinct spectral components (multiple basic frequencies), exhibiting distinct dominance in specific leads. The annotations of ventricular and atrial activities, ventricular and atrial trigger, as well as ratio between atrial and ventricular rates were greatly facilitated. The time diagram of ventricular and atrial triggers provides an additional view on rhythm disturbances.

CONCLUSIONS: The AF-Toolbox that is currently developed for clinical applications has the potential of reliably detecting and classifying AF, as well as to correctly describe atrioventricular conduction, propagation blocks and/or ectopic beats. Based on the results obtained, a first industrial prototype has been built, which will be used to assess its performance in a routine clinical environment. The availability of this tool will facilitate the search for meaningful signal features for identifying the source of AF in individual patients.