M.Sc. Stephanie Appel

B.Sc. Carl Enslin

Motivation
Simulating cardiac electrical activity is a complex and computationally intensive task.To address this, a hybrid model called the DREAM (Diffusion-Reaction Eikonal Alternant Model) was developed. It alternates between a cyclic fast iterative method solving the anisotropic eikonal equation and an approximation of a computationally intensive reaction-diffusion (RD) model, allowing for the simulation of complex arrhythmic conduction patterns like reentries at a much lower cost. A limitation, however, is its simplified, fixed-template approximation for the diffusion current, which struggles to accurately represent complex wavefronts.

This project aims to advance the DREAM model. Building on previous work to characterize and optimize the DREAM, optimized parameter sets have been identified that achieved up to a 40% reduction in simulation error and a 7% decrease in computation time in specific 2D scenarios. In parallel, a novel neural network model, adaINR (adaptive Implicit Neural Representation), is under development as a proof-of-concept to provide a more physiologically, adaptive diffusion current approximation. This model has shown promise in accurately representing diverse diffusion current morphologies while being computationally efficient.

Project Description
This project focuses on two main objectives to advance the DREAM model for fast and accurate cardiac simulations. The first goal is to validate the robustness and physiological accuracy of the optimized DREAM parameter sets in complex atrial models. This involves running simulations on realistic 3D meshes that account for anatomical and physiological heterogeneities. The project aims to characterize the DREAM's performance by analyzing how anatomical heterogeneities and pathological conditions impact conduction patterns and reentry dynamics.
The main part of the project involves the integration and evaluation of a machine learning-based approach to enhance DREAM’s core functionalities. This includes implementing the adaINR model into the C++-based openCARP DREAM framework. Then, the new hybrid adaINR-DREAM should be validated in performance and accuracy against the conventional DREAM and full monodomain simulations.