Before this program could be developed, there were several problems to overcome. First, the original NLM data provided no segmentation of the anatomical structures. Second, the sheer size of the NLM photo data was a concern. Third, compared to polygon-based rendering systems, the state of the art in volume rendering is rather infantile; in fact today it is yet considered an emerging technology. Last, but not least, the maturity and capability of the PC and Macintosh platforms to handle massive amounts of data and perform necessary computations was in question. The following paragraphs will explore how these problems were overcome to produce Virtual Human, the world's first three-dimensional anatomy atlas.

     GSM commissioned and funded Dr. Victor Spitzer's lab at the Center for Human Simulation at the University of Colorado to provide the first comprehensive segmentation of the Visible Human Male. A team of software engineers and anatomists tackled this formidable job over an eighteen-month period using proprietary software combining algorithmic processes with manual editing.

     The segmentation was performed on a cropped version of the photo images from the original NLM data. The existing data set was cropped from 2048 x 1216 to 1760 x 1024 to eliminate extraneous data, and perfectly registered. The cropped set of images is marketed by GSM as The Complete Visible Human Male Version 2.0.

     The segmentation effort provided information about the anatomical structures at more than 3.3 billion points in 3D space. Because there are more than 256 structures identified, and the data files are 16 bits in depth, the size of the data set is approximately 7 GB. GSM is the exclusive distributor of the segmented and classified data set, which identifies more than 1200 anatomical structures, and has been actively marketing this product in 1998.

     Several entities have already licensed the data for commercial and research projects. GSM has also recently announced an academic site licensing program which provides a considerable discount and wide distribution rights for noncommercial use at academic institutions.

     This segmented data set is the most accurate and complete segmentation of the NLM data in existence throughout the world. Recently, an update to the segmentation has completed, and is in processing for release. In this version the total number of structures segmented and classified is more than 1400. The process of updating and refining this segmentation is an ongoing effort.

     The future for the segmented and classified data lies in the imagination of researchers and entrepreneurs. There are clear applications for the data in simulation, education, reference, quantitative studies, and even art. While actively engaged in the licensing effort, GSM, as a medical publisher, created the first end-user product utilizing the data in order to both visualize the extraordinary data set and provide an unparalleled experience in exploring human anatomy.

     Virtual Human, the first truly three-dimensional anatomy software, performs live volume rendering of the photographic and segmented VHM data sets. Two versions of the program were produced. The Atlas version is targeted for medical education and professional reference. Physicians, medical students, and other healthcare professionals will find using the program to be a unique experience. The Studio version is designed for anatomy professors, medical researchers, multimedia producers, medical illustrators, and other graphics professionals. The program can produce photorealistic renderings and animations of publication quality. More details on the software's capability will be provided later in this paper.

     The first significant challenge in the development of Virtual Human was to deal with the volume of the data to be processed in order to provide a workable application. At this juncture, there existed approximately 7 GB of segmentation data and 10 GB of photo image data which would have to be processed on the target platforms.

     With this in mind, GSM employed a multiple-resolution scheme in which the data was resampled at three smaller resolutions. In the Atlas version, three resolutions are provided, with all four resolutions (including Full) implemented in the Studio version. In fact, this scheme is one of the chief reasons that bringing this large data set to the PC platform was possible. Based on the user's request for a set of specific anatomical structures, the program calculates the smallest possible volume (region of interest), and, based on available memory, loads the appropriate resolution. The user can exercise explicit control over the resolution to preview renderings or to create final publication-ready images. The sizes of the segmented and classified data set in the various resolutions are outlined in Table 1.

     Through an exhaustive search and evaluation period GSM put several existing volume rendering engines through a rigorous test with this massive data set. Fortunately, we discovered an excellent engine produced by Fortner Software of Sterling, VA, which was capable of handling this data set on PC and Mac platforms, and were able to forge a licensing agreement. In developing Virtual Human, we selected those parts of the rendering engine that met our requirements exactly, and when portions did not, we developed interfaces and code that maximized its capabilities for our application. Finally, our attention turned to tuning the software to work within the unique data handling and memory-allocation routines of the target platforms. Through this process, we have pushed these platforms to their limits, allowing the Visible Human Male to be presented on platforms familiar to a majority of the population. The results, briefly described below, have been more than satisfactory.

     Interacting with the three-dimensional scene, one can rotate the anatomy, move closer to or farther from it, customize the lighting around it, and probe or label structures within it. Slices and geometrical cutouts can be created from the data, and individual structures can be highlighted with colors or made semi-transparent. Anatomical shapes and relationships are seen in ways not previously possible.

Figure 1 shows an image rendered at high resolution. The image illustrates variable opacity of individual structures and coloring effects for highlighting.

Figure 2 was rendered at high resolution and employs some of the cutout techniques available in the program.

     Unlike the Atlas version, the Studio version contains the anatomical datasets at full resolution for maximum rotation, flyby, sequential slicing, and other motion effects. Additionally, Studio contains professional accuracy and visual impact. It also contains animation tools with which you create digital movies demonstrating annotation features for publication quality labeling.

Figure 3 was rendered at full resolution and also shows the use of structure opacity and automatic labeling of structures.

Movie 1 shows the animation production capabilities of Virtual Human and is a Quick Time™ animation created using high resolution images.

Figure 4 was rendered at high resolution, and shows the use of structure opacity, image orientation, automatic labeling, and lighting. Compare with Netter's Plate 400.

     Although Virtual Human has a steeper learning curve than the average multimedia educational tool, it brings with it a level of power and insight not available previously. Furthermore, as the inevitable progression of processor speeds continues, the advent of real-time volume rendering will enable drastic simplifications of the program's interface to enrich the "virtual reality" experience. For more information, please see http://www.gsm.com/products/virtual.htm.