High Performance Computer Simulations for the Study of Biological Function in 3D Heart Models Incorporating Fibre Orientation and Realistic Geometry at Para-Cellular Resolution MO Bernabeu, MJ Bishop, J Pitt-Francis, DJ Gavaghan, V Grau, B Rodr´ ıguez Computing Laboratory, University of Oxford, Oxford, UK Abstract Information regarding the propagation dynamics of in- tramural electrical wavefronts in the 3D volume of the ventricles is critical in gaining a better understanding of normal and pathological cardiac function. However, in- vestigation into such phenomena using cardiac modelling has so far been impaired due to limitations in the struc- tural details included in current cardiac computational models. Here, we describe the technological pipeline re- quired for the construction of realistic, highly-detailed and personalised whole ventricular models directly from high- resolution MR images and their use within a reliable, fully tested cardiac simulation software (Chaste). Simulations of cardiac propagation in the structurally-detailed model are presented that reveal the importance of complex struc- tural geometry, fibre orientation, blood vessels and other heterogeneities in propagation of activation wavefronts through the ventricular volume. The tools and techniques presented in this study are expected to be key in the devel- opment and application of the next generation of cardiac models. 1. Introduction Knowledge of the specific dynamics of intramural prop- agating electrical activation wavefronts throughout the heart is critical in gaining a better understanding of car- diac function in both healthy and pathological conditions. However, the way in which wavefronts interact with, and propagate through, heterogeneous regions of tissue has not been thoroughly assessed due to limitations in the detail and complexity of current whole ventricular models. The goal of this study is to develop the techniques to build re- alistic, highly-detailed and personalised whole ventricular models directly from high-resolution MR images for use within a reliable, fully tested cardiac simulation software. The software and models developed here are used to sim- ulate electrical propagation in the 3D volume of the rab- bit ventricles, with unprecedented resolution in myocardial structure. To achieve this goal, a realistic anatomically-detailed finite element ventricular model was constructed directly from a high-resolution (voxel size ≈ 25µm 3 ), 3D rabbit MR data set. The images were segmented using a combi- nation of level-set techniques and used to generate tetra- hedral meshes. The model includes detailed structural in- formation including blood vessels, papillary muscles, tra- beculations and a representation of realistic fibre orienta- tion. Propagation of electrical activation within the realis- tic rabbit ventricular model was simulated by solving the monodomain equations using Chaste, a novel biological simulation environment. Chaste has been developed using state-of-the-art numerical and computational techniques, as well as professional software engineering practices, and provides a rigorously-tested library of computational biol- ogy software. 2. Development of computation model 2.1. MR data acquisition and processing MR scans were performed on a fixed and agar embed- ded female rabbit heart (≈ 1.2kg) using an 11.7 T(500 MHz) MR system. For specific details of the MR sys- tem and scan protocols used see [1]. Scans were acquired with an in-plane resolution of 26.4 µm × 26.4 µm and out-of-plane resolution of 24.4 µm producing a MR im- age stack containing 1024 × 1024 × 2048 voxels. Due to computational memory restraints, however, this was down- sampled once by a factor of 2 using Matlab (The Math- Works, Inc.) to produce a more managable data set con- taining 512 × 512 × 1024 voxels. Fig. 1a shows a cross- section of the MR data set taken along the short-axis of the heart viewed using the freely available software Seg3D (software.sci.utah.edu/). 2.2. Segmentation of MR data set Segmentation is the critical first stage in extracting ge- ometrical information from the MR data set to build a ISSN 0276-6574 721 Computers in Cardiology 2008;35:721-724.