Pranav Bhagirath

160 Chapter 8 DISCUSSION In this study the feasibility of computing BVP for non-invasive imaging of cardiac excitation is illustrated. So far, volume potentials have been considered to be of limited value. However, computing BVP using the FEM is efficient if the number of bounding surfaces between organs taken into account is high. In addition, using the FEM rather than the BEM enables the incorporation of different anisotropies, local tissue characteristics and sigma gradients over different regions. BVP may be used to gain a better insight in the genesis of BSP and sources of local inaccuracies. Figure 6 suggests that the computed BVP hardly depend on themesh cell characteristic length for a ratio as big as 1:3. However, the tank experiment indicates that there are geometries where the mesh size does have a significant influence on the outcome. The FEM could be used to understand which aspects of the geometry caused the large misrepresentation of the 4V equipotential plane. By visualizing the large potential gradient near the electrodes and the small potential gradient in the middle of the box, the FEM contributed to understanding the inaccuracy of the computed position of the 4V equipotential plane. Computation of BVP Froma computational standpoint, computing a potential fieldbymeans of the FEM is not a problem. The computation time of a 240,000 points potential field was approximately 3 minutes. Graphics processors with hundreds of computation units combined with quadcore main CPU’s are becoming available at consumer prices. Generally available FEM packages are able to benefit from this just by setting some parameters and no custom coding. While the performance gain by computing the FEM in parallel may easily be tenfold. The FEniCS package has been selected because this tool has a lot of mindshare, a vivid user community and excellent documentation. Meshing Generating high quality meshes and solving differential equations have been part of engineering disciplines for decades. In many articles on biomedical computing, generation of a computational mesh is taken for granted. Early experiments revealed that generation of a surface mesh from a labelled voxel set often leads to irregular meshes with topological errors that are hard to repair. Several groups have developed their own meshing software and others have proposed to do away with meshing altogether [16-18]. However, the meshing itself does not appear to be the problem.