Investigating termination mechanisms of chaotic spiral and scroll wave dynamicsunderlying cardiac by using hypothesis-driven and AI-driven terminationapproaches

点击翻译按钮获取中文摘要

Life-threatening cardiac arrhythmias, such as ventricular fibrillation, are a major cause of morbidity and mortality worldwide. The self-organized spatio-temporal dynamics observed during arrhythmias is determined by vortex-like rotating excitation waves and their nonlinear interaction with each other. Since the currently used conventional treatment method (the delivery of a high-energy electrical defibrillation shock) is associated with severe side effects, research is being conducted into the development of alternative treatment options, particularly low-energy defibrillation methods. Numerical simulations can be used to investigate relevant mechanisms of low-energy defibrillation and to develop new effective pulse sequences in detail, and thus contributing in this way to minimize the number of preclinical animal studies. In this project, we want to follow two approaches on how to develop and investigate efficient control strategies for chaotic spiral/scroll wave dynamics underlying cardiac arrhythmias: A. Hypothesis-driven termination approaches, where pulse sequences are motivated and developed based on underlying (physical) mechanisms and B. AI-driven termination approaches, where concepts and algorithms from the field of Reinforcement Learning (RL) will be used. In the first part of the project (A) we want to investigate how hypothesis-driven termination approaches which we previously developed in two-dimensional and simplified numerical simulations perform in three-dimensional and realistic heart geometries. In detail, we want to study the influence of features like a non-trivial geometry, a realistic conductivity anisotropy and 3D properties of the dynamics (e.g. filament tension of scroll waves) as well as the interaction of these features with regard to the termination rate. In the second part of the project (B), algorithms from the field of reinforcement learning will be used: An agent (an artificial neural network) interacts with the system/the dynamics to be controlled (chaotic wave dynamics) by the application of spatially localized perturbations. Guided by rewards or punishments based on the effects of the perturbations, the goal is to develop effective perturbation patterns/pulse sequences that are free of mechanistic assumptions. Finally, we can compare AI-driven control strategies with the current hypothesis-driven approaches, in order to investigate what the similarities and differences between both concepts are. In collaboration with our experimental partners we will discuss how the obtained insights can be investigated and verified in experimental setups and how to incorporate the gained knowledge into the development of novel low-energy defibrillation therapies.