One fundamental question that this investigation sought to address was whether a high-resolution, 3D, laser-scanned, digital model would be useful to paleontologists. Certainly a digital model allows for a new method of illustration. But in some cases one must consider whether a digital model represents a viable replacement for a physical specimen or cast?

Several areas of the 3-million polygon digital model were examined and compared to the specimen. For example, in the case of the basisphenoid-basioccipital presented here, areas of interest to a researcher examining a mosasaur braincase would be blood supply to the brain and the cranial nervous system (Russell 1967). Obviously, soft tissues are rarely preserved (Fastovsky and Weishampel 1996), but cranial nerves often left pathways and foramens throughout the brain case as they traced their way from the extremities of the animal (e.g., sensory organs) to the various reception sites in the brain (Hildebrand 1988). These foramens can be essential to researchers because they indicate the likely position and structure of the brain (Hildebrand 1988) and can provide important characters for a phylogenetic analysis (Bell 1997). As a result of alteration during the fossilization process, the illustrated basisphenoid-basioccipital (Figure 3 and Figure 5) shows some deformation and compression along the lateral margins, slightly obscuring one of the foramen of the abducens nerve (cranial nerve VI; Fig. 8). However, the structure is clearly visible, even scanned at 100 Ám resolution.

Blood circulation to the brain also provides important clues to phylogenetic relationships, with the vidian canal likely containing the internal carotid artery (Russell 1967; seen clearly in Fig. 7). The branch of the internal carotid artery leaves the vidian canal and passes medially through the basisphenoid (Fig. 8). In some mosasaur clades the internal Carotid Artery became significantly enlarged so it is of phylogenetic importance to reconstruct this vessel.

Comparison of 50 Ám and 100 Ám scan resolution

Would a 100 Ám scanning resolution be sufficient to produce an accurate digital model for paleontological study, or is a 50 Ám scanning resolution a requirement? The answer to this rhetorical question is important because scanning at higher resolution generates a higher polygon count model and larger file sizes and requires a significantly more powerful computer to manipulate. As discussed above, whether a 50 Ám or 100 Ám resolution scan is sufficient for research purposes will depend on the individual researcher's interests. As an example, a small area of the basal tuber (Fig. 10) scanned at 50 Ám was compared to the same region at a 100 Ám scan. The basal tuber of the basisphenoid is ridged and pitted, likely from a rich supply of blood through a network of blood vessels supplying a cartilage covering (Russell 1967). When the pits and ridges of the basal tuber are magnified (Fig. 13), the internal structure of the ridges are clearly seen at 50 Ám scanning resolution. Unfortunately at 100 Ám scanning resolution the pits and ridges are faceted at the equivalent magnification. Clearly, if a researcher is interested in the internal structures of small features such as the basal tuber, or other structures smaller than 100 Ám, a scan resolution of 50 Ám is required. However, it is likely that, for a majority of researchers and for most applications, a 100 Ám scan resolution would be satisfactory because it supports artifact-free magnifications of up to 5x.

As laser scanning technology matures, as the process becomes increasingly automated, and as personal computers become faster, cheaper, and more common, scans at 50 Ám resolution will become more practical permitting more information to be presented in a 3D model.

Polygon Reduction

While computer speed has increased dramatically in recent years (Moore 1997), most desktop computers have yet to achieve performance levels suitable for manipulating 3D digital models comprised of hundred of thousands of polygons in real time. Modern Silicon Graphics workstations allow real-time manipulation of digital models composed of millions of polygons. Unfortunately, the prohibitive cost of these high-end graphics computers makes them unavailable to most researchers. To permit most readers access to a version of this complex digital model that can be manipulated, it was necessary to reduce the number of polygons to a more manageable number. 'Manageable numbers' of polygons for desktop computers are dependent on the speed of the processor, the speed of the video card, the amount of video memory, the amount of random access memory available, and the file type. At this time 50,000 to 800,000 polygons seems to be the limit for the most recent desktop machines (e.g., Intel Pentium III, AMD Athalon, and Motorola G3/G4 processors).

Consideration of the differences between 3D file types is critical when assessing the ability of desktop computers to display a complex model composed of many polygons. Some file types support texture maps and multiple light sources and allow for compact file sizes, but are not widely viewable. In contrast, other file types do not support texture maps and are inefficient, but are well-supported and easily read by a variety of operating systems (Macintosh, PC, and Unix). An example is InnovMetric's Polygon file type versus VRML. With the Intel Pentium III computer, used for this research, with 384 MB of RAM, an InnovMetric polygon file composed of 800,000 polygons is easily manipulated in near real-time, while the same model must be reduced to 50,000 polygons to achieve the same response when viewed using VRML. The benefit of using VRML though is that it is a standard file type and can be read by many different types of software. These VRML files will also likely still be readable for the foreseeable future because the files are text based. For these reasons, VRML was chosen as the preferred file type for this study even though it is not nearly as efficient as an InnvoMetric polygon file format. Nevertheless, it should be noted that VRMLs inefficiency required that the number of polygons be reduced from 3 million polygons to 50,000 polygons; a reduction of 98%. While this allows for the model to be displayed on fast desktop computers, an assessment must be made as to whether any paleontological value remains in such a substantially reduced model. An algorithm present in PolyWorks/Modeler that removes redundant information carried out the process by which the number of polygons was reduced. Results suggest that if the 50,000 polygon digital model is displayed at its original (1x) size, no apparent difference can be noted between the physical specimen and the digital model (Fig. 3). However, as the digital model is scaled upwards (simulating examination of the physical specimen at increased magnification), differences quickly become apparent. At a scale equivalent to 1.5x and upwards, the fine details and ridges apparent in magnified views of the physical specimen are either obscured or absent (Fig. 14). Thus, polygon-reduced models only have limited value to researchers, for example in roles such as informal consultation with colleagues or popular science.

Polygon-reduced models potentially do have a role in paleontological research in other applications such as virtual reconstruction though. For example, if the five remaining braincase bones (right quadrate, right and left prootic, parietal and supraoccipital) of the juvenile Tylosaurus sp. were also scanned at 100 Ám resolution and then assembled to create a virtual reconstruction of the brain case, the resultant model would likely comprise over 12 million polygons and would be in excess of 200 MB. In order to manipulate such a reconstruction, the number of polygons would have to be reduced by an order of magnitude or more depending on the researcher's hardware and software. The resultant reduced model would still be useful for various applications including estimation of brain morphology and volume. One can imagine the number of polygons present in a complete whole skeleton reconstruction though; it could conceivably number in the billions leaving this application in the realm of science fiction for the next few years.