Artifact vs. Anatomy: Dealing with Conflict of Geometric Modeling Descriptions

In applications ranging from design of customized vehicle interiors to virtual testing of biomedical devices, the processes of modeling, design and analysis involve the simultaneous treatment of artifacts (i.e., parts designed by humans) and anatomical structures. An inherent conflict arises because the geometric descriptions are completely different. Artifact descriptions are typically the output of computer-aided design (CAD) software and consist of a collection of parametric patches that comprise the boundary of the artifact. In stark contrast, the native description of an anatomical structure typically consists of an image stack obtained using a volumetric scanning technology such as computed tomography (CT) or magnetic resonance imaging (MRI). Current practice for simultaneously dealing with both categories of entities involves working primarily in the world of CAD. The scanner data must be processed to determine segmentation (i.e., to decide which voxel belongs to each object of interest), and then the segmented results are converted to traditional boundary representation (b-rep) CAD models. However, the CAD models consist of such large numbers of triangles (often on the order of 10⁶ or 10⁷) that performing standard CAD operations becomes problematic. Most traditional CAD systems are designed to deal with environments composed of at most 1000's of individual objects with tens of thousands of individual surfaces (beyond which many CAD systems- performance is so slow as to have effectively failed).The question then arises whether forcing everything into the traditional CAD environment is really the best approach. If anatomical structures have more complicated geometry than CAD artifacts, does it make sense to work with anatomy in the CAD environment? In this paper we present a modeling representation whose basic data is much closer to the natural representation of scanned objects in that the representation is based on a 3D grid of data. While the raw scan data is too noisy to be directly useful, the grid of signed distance values that are obtained by recently developed segmentation algorithms is well behaved and amenable to accurate interpolation by wavelets. Thus we present a modeling approach that employs wavelets to interpolate signed distance values resulting in useful function-based representations (f-reps). We present examples to show how this approach produces models of scanned structures, provides for import of artifacts, and supports simultaneous modeling of artifacts and anatomy.