Our goal was to create a simple digital representation of the human brain
suitable for tomographic image simulation. The model needed to have high
resolution (e.g., high enough to simulate 1mm3 MRI data sets) and must
contain properly model PVEs so that classification algorithms
can be tested. Furthermore, the model must be anatomically accurate and as
realistic as possible to have greater value than more simply-shaped
phantoms. Most importantly, in order to be realistic, the model should be
completely 3D.
A voxel representation was selected since smooth transitions can be more easily modeled than, for instance, polyhedral models. Other reasons include: the availability of 1) a high-resolution low-noise data set [15] that could be used as an anatomical guide, 2) automatic classification procedures to correctly label the majority of voxels, and 3) 3D visualization/editing tools to facilitate manual correction during the voxel-labelling phase, i.e., attachment of an identifying tissue type to each image voxel. The following sections describe phantom construction in more detail.