EXTRACTING VIRTUAL BONES FROM CT DATA

One of the two most detailed and expensive techniques of creating 'virtual' bones is high-resolution computer tomography (HRCT, short CT) scanning specimens. This allows maximum resolution, far higher than required for most uses, similar to high resolution laser scans of individual bones. The former technique has the advantage of allowing the study of internal structures and does not suffer from 'blind spots', as X-rays penetrate the material. Even surfaces completely blocked from view such as deep cavities and recesses on skulls are faithfully reproduced in the virtual bones. Virtual bones from both methods can be assembled into virtual skeletons either simply based on their own shapes, much as it is possible for real bones. Drawings, photographs, or measurements of mounted skeletons can be of help, but are rarely required, since the high-resolution virtual bones provided by both methods contain all the information needed for assembly. One drawback of these methods is the relatively large file size. Both Rhinoceros® and the Geomagic® program suite offer options for reducing the number of polygons in each mesh, reducing the file size proportionately. The latter program offers the additional option of preserving the outside contours better and thus should be preferred. While reducing the mesh number decreases file size, the resulting virtual bones lose accuracy, and the reduction should not be taken too far. On average, a reduction to 2.5 to 10% is the maximum tolerable, depending on bone size and shape. Delicate structures may start losing shape at 20% reduction already (i.e., 80% of the original size).

For data extraction, the files of one scan are loaded into AMIRA 3.11®. Then, a 'LabelVoxel' module is created and applied to the data. Here, up to four different areas of density can be defined. A histogram is helpful for interpreting the data and deciding where to set the borders. It is possible, e.g., to remove or include plaster fillings by choosing different settings. Now, an 'OrthoSlice' module can be created to view cross sections. In order to keep the computing time and memory requirements low, the re-labeled data should be cropped to contain no unnecessary space, e.g., empty space under or above the bones. Large bones should be cropped out so that each bone is treated separately. Since the original data is still present in unaltered form, after extraction of the first bone it can simply be 'labeled' again and the next bone treated. To each cropped set of labeled data, a 'SurfaceGen' module is attached and executed. This creates a polymesh surface, which can be saved as a number of different formats, e.g., ASCII stereolithography (*.stl). The resulting files are highly detailed and accordingly huge. A longbone can easily have 10 million polygons and exceed 1 GB in file size. To reduce the size, it is useful to load the files into Rhinoceros® and re-save them as binary STL files, which have a significantly smaller size without any data loss. Reducing the number of polygons, on the other hand, results in a less accurate representation of the surface. Usually a reduction to 20% is hardly noticeable to the human eye if a bone is displayed at full-screen size. Therefore, a slight to generous reduction may be acceptable depending on the planned use of the data. As mentioned above, this is best done in Geomagic®, as this program has an option to 'preserve edges', guaranteeing a minimum of shape change during polygon reduction. AMIRA 3.11® also offers this option, here called 'Simplifier'. 'Preserve slice structure' is the equivalent to the edge preservation option in Geomagic®.