Next: Color Segmentation Up: Title Page Previous: Full Text Index Index: Full Text Index Contents: Conference Page

Introduction

Rapid advances in computer graphics, over the last decade, have triggered new fields which deal with visualization of complex and large data sets. The National Library of Medicine introduced the Visible Human Project [1] [26], and the participants in the project are working toward producing a complete library of high-resolution color 3D representations of an adult male and a female cadaver. The Visible Human data consists of color 2D slides that allow for the representation of details which had been all but invisible in more traditional data sets. However, in order to make the best use of the data, accurate and efficient methods must be developed to identify structures within the individual 2D slices. Outlines of the segmented structures can then be used to extract 3D voxel-based models from the 2D data, and the 3D surface-defined wrappings. These models must not only be accurate enough to allow for labeling of all important and visible anatomical structures, but must also be efficient enough to be usable in interactive applications, such as the computer based anatomy curriculum being developed by the Vesalius(TM) Project [*] at Columbia University [2] [10] [11] [18] [27] [28] [29] [30].

The goal of our research is to address the following issues:

(a). Registration: The Visible Human (VH) data set which consists of 2D digitized slices is not always perfectly aligned [15]. The alignment should be done by following either the fiducial rodes which are preserved in the original data or internal (anatomical) structures which pass through a large number of slices (e.g. aorta).

(b). Color Segmentation: Generating outlines for the structures, which falls under the umbrella of image segmentation, stands out as one of the most challenging and vital phases of the project, primarily because the results of all subsequent steps in the process depend on the quality of the initial 2D segmentation.We must be aware that segmentation of some of anatomical structures will always require a manual intervention. Also, a data base of segmented anatomical entities should be built, from which a segmentation of new, more complex anatomical structures can be derived.

(c). Surface Reconstruction: To obtain a color meshed surfaces from the voxel-based reconstruction of the anatomical structures, a modification of the marching cubes algorithm [13] [17] where the mesh is textured by "dipping" its vertices in the original color data can be used. Since the 3D surface-based models reconstructed from VH slices reveal "jagginess" after the marching cubes algorithm is applied, a smoothing algorithm must be used to correct the problem.

(d). Multi-resolution Mesh Representations: The color models of anatomical structures are highly detailed and maintain a level of realism that should satisfy anatomists who teach anatomy classes. But the high-precision surface triangulations, in which RGB colors are computed for each triangulation vertex, cannot be manipulated in real time on an arbitrary platform and transmitted efficiently. Also, the full complexity of such models is not always required, therefore simpler and less expensive version of such complex mesh models can be derived. The mesh-reduction algorithm should be automated and produce a user specified hierarchy of multi-resolution meshes where, in addition, the original texture would be preserved.

(e). Volume Rendering: Some applications require volume-rendered scenes where a number of structures are displayed together and immersed in a ``gel-like'' transparent voxel-based environment. We face a challenging task of implementing a color segmentation algorithm which would be able to process very large set of voxels. The existing volume rendering algorithms must be re-visited to accommodate the size of the data set. Otherwise, a high compression of the voxel data must be used.

(f) Interactive Browsing: An alternative to presenting a complex scene populated with a number of 3D models, generated by a volume rendering algorithm, is to use their surface-based representations and allow browsing through such a scene using ``peeling'' via animated transparency. We would like to focus on developing an interactive browsing of complex 3D scenes (e.g. 3D body maps) representing anatomical regions, in conjunction with a query language and a descriptive and explanatory text, derived from an ontology [28].

We are interested in refining the available technology in order to obtain the high photographic quality, smooth models, from the Visible Human data, which can be manipulated in real time on any platform and transmitted efficiently [7].

We collaborate closely with the medical faculty at Columbia to validate the outcome of the visualization. Because of the unusual complexity of the color anatomical images we must constantly consult anatomists and other medical specialists.

On the broadest level the Vesalius group is seeking to develop a computer-based learning environment for the medical and dental students at the Columbia University College of Physicians and Surgeons and the School of Dental and Oral Surgery. This group of 225 students are educated together during the first two years of their training. The Vesalius group is introducing technology into the curriculum in cases where information technology can make a difference, where the book, chalk, overhead transparencies, slides, etc., are simply inadequate to the tasks of teaching and learning. The group is creating an organized, consistent, and reliable information resource that links text and image as directed by faculty for teaching or called for by students for review and study.

In the next few sections we describe, in more details, the tools which are being developed for the Vesalius Project.

(*) Trademark held by Columbia University. The Vesalius Project is named after Andreas Vesalius, a sixteenth century anatomist whose work laid the foundations for all subsequent anatomical research. The Columbia University Health Sciences Library owns several first editions of his work.

Next: Color Segmentation Up: Title Page Previous: Full Text Index Index: Full Text Index Contents: Conference Page