Session P91.4

Semi-Automatic Enhancement of Atrial Models to Include Atrial Architecture and Patient-Specific Data: For Biophysical Simulations

BD Flores-Hermosillo*

University of Hull
Hull, UK

Adjusting a Reaction-Diffusion formulation for patient-specific simulations isn’t simple; a feedback method must be implemented. Additionally, the underlying anatomy representation is often oversimplified, despite its large influence over the simulation results.
The simulation domains (i.e. anatomy and physiology) of biophysical models are tightly intertwined. The disagreement observed between real and synthetic electrical mapping data has complex origins; pinning the causes to a domain is not quite possible yet.
An atrial model including microscopic muscle architecture would augment the correlation of results. Furthermore, it would also provide means to de-couple the simulation; monitoring discrepancy factors such as preferential conductivity, continuity, and heterogeneities. Setting these model requirements and providing a tool which facilitates its creation are the objectives of this work.
The initial input is any mesh (tetra or hexahedral) of the atria, obtained from a segmented cardiac MRI dataset. The mesh can be structured or unstructured, adaptive or not. An electric grid is then embedded in the mesh as follows: each geometric element represents a myocyte, the basic simulation element; a grid node on its centroid. Full connectivity by gap junctions to all its adjacent elements is initially assumed.
Segmentation into a hierarchy of muscle bundles, following histology data, is the first step. Next, inter-bundle connectivity is specified in different domains of the total volume of the bundle and in a per-connecting-bundle basis; thus, introducing discontinuities using both, known parameters and a “random” process. Lastly, fiber orientation (FO) is assigned per bundle by fitting control points to bundle axis. These automatic processes require little use intervention.
The results can then be manually tuned based on observations from patient data: various sample images of MR and DT. FO design requires special attention in order to include features like fiber rotation through tissue layers.
The resulting output model includes features useful for simulation such as “severing” gap junctions, altering basic myocyte functionality, control of local and regional model properties. The potential is on hand for creating special bundles: e.g. conduction or isolation paths, or a SAN bundle, and manipulating their physical extent and biological behavior. The model is designed for FV solvers.
The bulk job is done automatically and although the manual processes are time consuming, these are nevertheless intuitive and facilitated by a friendly GUI. The algorithm has been tested on the aforementioned mesh topologies and it has proven to be effective. Intermediate stages can be saved, retrieved and edited to alter the final product; thus providing flexible resources for experimentation.
An added value is the design, documentation, and coding of the software; good practices for systems engineering were employed such as using UML and OOD.

(Abstract Control Number: 157)