Mechanisms of stochastic onset and termination of atrial fibrillation\n episodes: Insights using a cellular automaton model

Yen Ting Lin, Eugene Ty Chang, Julie Eatock, Tobias Galla, Richard H. ClaytonarXiv 2016 · DOI · arXiv

Abstract

Mathematical models of cardiac electrical excitation are increasingly complex, yet their computational cost often prohibits the study of long-term behaviors like atrial fibrillation (AF). This study implements a Moe-type cellular automaton on a topologically correct 3D surface geometry of the left atrium to investigate the stochastic initiation and spontaneous termination of AF. By generating over $10^5$ sample paths, the researchers quantified transition rates and identified a "memory" effect where the stability of AF episodes depends on the duration of the initial trigger.

Introduction

Atrial fibrillation (AF) is a prevalent cardiac arrhythmia characterized by rapid and irregular electrical activity in the atria, significantly increasing the risk of stroke and heart failure. While biophysically detailed models provide deep insights into cellular mechanisms, they are computationally prohibitive for simulating the long-term, stochastic nature of AF episodes, which can range from seconds to months. To bridge this gap, this research utilizes a simplified cellular automaton (CA) model that maintains essential spatial and physiological features while allowing for the massive parallel simulation required for statistical inference.

Methods: A Computationally Efficient Atrial Model

The study employs a Moe-type cellular automaton, which discretizes atrial tissue into nodes with states such as "at rest," "activated," or "refractory." Unlike traditional 2D grid models, this CA is implemented on a realistic 3D geometry.

The 3D spherical geometry of the left atrium model, showing the locations of the pulmonary veins (PV) and mitral valve (MV), with red dots indicating fibrosis.

The model incorporates several critical physiological factors:

  1. Realistic Geometry: An icosahedral mesh of approximately 10,000 nodes is mapped onto a sphere representing the left atrium, featuring anatomically accurate placements for the four pulmonary veins (LSPV, LIPV, RSPV, RIPV) and the mitral valve (MV).
  2. Fibrosis and Restitution: Randomly distributed "dead" nodes simulate fibrosis, while a restitution rule adjusts a node's recovery time based on the timing of its previous activation.
  3. Stochastic Triggers: AF is initiated via random bursts of high-frequency activity near the pulmonary veins, mimicking clinical observations of ectopic triggers.

The primary advantage of this approach is speed; the simulation runs 16 times faster than real-time, enabling the generation of up to $10^6$ sample paths for rigorous probabilistic analysis.

Results and Key Insights

The large-scale simulations revealed several fundamental properties of AF dynamics that are difficult to observe in shorter, more detailed simulations.

The "Memory" of AF Episodes

A significant discovery is that AF episodes exhibit a form of "memory." The stability of a resulting AF rhythm is strongly correlated with the duration of the trigger that initiated it. Episodes triggered by short bursts (e.g., 1 second) tend to be unstable and terminate spontaneously. Conversely, longer triggers (e.g., 150 seconds) lead to highly stable AF patterns that are much more resistant to termination.

Simulated activation patterns and the probability of reentry as a function of PV burst duration and other model parameters.

Stochastic Initiation and the "Sweet Spot"

The probability of AF initiation does not increase linearly with the trigger frequency. Instead, there is a non-monotonic relationship: an intermediate frequency range exists where the tissue is most susceptible to reentry. If the trigger is too slow, the tissue recovers fully between beats; if too fast, the signals often fail to propagate due to the refractory period, preventing the formation of sustained rotors.

Quantifying Transition Rates

By fitting the simulated data to a two-state Markov model, the authors calculated concrete rates for initiation ($r_1$) and termination ($r_2$). This allows for the propagation of microscopic uncertainties—such as the specific spatial arrangement of fibrosis—into macroscopic predictions about the likelihood of a patient remaining in AF over long periods.

Discussion and Future Directions

While the spherical geometry is a simplification of the complex human atrium, this model provides a powerful "middle ground" for cardiac research. It allows for the exploration of parameter spaces that are currently inaccessible to full biophysical models.

Future work aims to incorporate more complex anatomical features, such as varying wall thickness and directional fibrosis (anisotropy). Furthermore, the transition rates derived from this model can be integrated into clinical care pathway models to better predict long-term patient outcomes and the efficacy of different therapeutic interventions.

Conclusion

This research demonstrates that simplified cellular automaton models, when combined with realistic geometries and large-scale statistical sampling, can uncover complex emergent behaviors in cardiac arrhythmias. The identification of AF "memory" and the quantification of stochastic transition rates represent significant steps toward personalized, long-term management of atrial fibrillation.

Summary updated

Abstract

Mathematical models of cardiac electrical excitation are increasingly\ncomplex, with multiscale models seeking to represent and bridge physiological\nbehaviours across temporal and spatial scales. The increasing complexity of\nthese models makes it computationally expensive to both evaluate long term (>60\nseconds) behaviour and determine sensitivity of model outputs to inputs. This\nis particularly relevant in models of atrial fibrillation (AF), where\nindividual episodes last from seconds to days, and inter-episode waiting times\ncan be minutes to months. Potential mechanisms of transition between sinus\nrhythm and AF have been identified but are not well understood, and it is\ndifficult to simulate AF for long periods of time using state-of-the-art\nmodels. In this study, we implemented a Moe-type cellular automaton on a novel,\ntopologically correct surface geometry of the left atrium. We used the model to\nsimulate stochastic initiation and spontaneous termination of AF, arising from\nbursts of spontaneous activation near pulmonary veins. The simplified\nrepresentation of atrial electrical activity reduced computational cost, and so\npermitted us to investigate AF mechanisms in a probabilistic setting. We\ncomputed large numbers (~10^5) of sample paths of the model, to infer\nstochastic initiation and termination rates of AF episodes using different\nmodel parameters. By generating statistical distributions of model outputs, we\ndemonstrated how to propagate uncertainties of inputs within our microscopic\nlevel model up to a macroscopic level. Lastly, we investigated spontaneous\ntermination in the model and found a complex dependence on its past AF\ntrajectory, the mechanism of which merits future investigation.\n
BibTeX
@misc{lin2016mechanisms,
  author = {Yen Ting Lin and Eugene TY Chang and Julie Eatock and Tobias Galla and Richard H. Clayton},
  title = {{Mechanisms of stochastic onset and termination of atrial fibrillation\textbackslash\{\}n episodes: Insights using a cellular automaton model}},
  year = {2016},
  eprint = {1612.03403},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1612.03403},
}
Figure