Predicting Termination of Atrial Fibrillation Based on Shape Analysis in Reconstructed State Space

Abstract

It is important to recognize and characterize conditions under which atrial fibrillation (AF) is likely to terminate spontaneously or be sustained. This leads to not only better understanding of onset and termination mechanisms of the arrhythmia but also improved treatment of sustained AF. In this paper, a novel methodology is proposed to characterize AF from a nonlinear point of view to predict its spontaneous termination. The state space of the electrocardiography (ECG) signal is firstly reconstructed and points on a given Poincare surface are studied. A Poincare surface is a hyperplane intersecting the coordinate axis of interest at a given position in the reconstructed state space. Then a regularity index, p is proposed to characterize the geometric structures of the points on Poincare surface based on the shape analysis. It is defined by calculating the relative number of similar pairs of points on the Poincare surface. Finally, a threshold is calculated as the value of p which minimizes the prediction error for the training data set to classify non-terminating and terminating AF.Here an AF database which includes a training set with known termination properties and two testing sets of Holter ECG recordings is studied in the experiments. Results show that with the feature proposed in this study, 90% of the testing set is correctly classified. It demonstrates that the algorithm has the ability to predict spontaneous termination of AF from Holter ECG recordings effectively.

Predicting the spontaneous termination of atrial fibrillation based on Poincare section in the electrocardiogram phase space

“…Thus, there have been a limited number of nonlinear studies in which the RPS domain structure was employed, such as in our methodology. 7,19,25…”

“…Phase space reconstruction in lower dimensions, equal to two, and reconstruction based on the full ECG rather than just its AA and utilizing new features are the main differences between our work and the research done by Sun and Wang in phase space with embedding dimension equal to 200. 25 From a representation perspective, Poincare sections with four or more dimensions are not useful because they cannot be viewed on a monitor or plotted on paper. Visualizing the Poincare section on a monitor or paper can benefit clinicians with a new tool which is complementary to reading the ECG on paper or a monitor.…”

“…The present study utilized a new method to quantify AF dynamics with different termination properties: N-AF, S-AF, and T-AF. Along with the presentation of a new rule for determining the coordinates of a Poincare line which was not perpendicular to the RPS axis as in previous studies, 25,40,41 a simple method was used to highlight the Poincare intersection in the region of the RPS related to AA which could be applied to a sample comprising a single lead with a small number of data and capture POS related to AA of the lead under consideration. In addition, different features for quantification of AF termination type, including numerical, statistical, and distribution-based features, were used.…”

“…52 A number of studies assumed that the Poincare plane is perpendicular to one of the RPS axes and used the correct classification rate as a criterion to select coordinates of the Poincare plane along that axis, adopting Poincare planes that resulted in the minimum classification error. 25,40,41…”

“…Yang proposed further in-depth investigations in this new framework 41 and Sun and Wang suggested introducing new features for multivariate prediction of AF. 25…”