Next: References Previous: Title Page Contents: Conference page

Using the Visible Human Dataset to Represent Part-whole Relations
in the Vesalius Navigational Ontology

Overview
(1) A navigational ontology.
(2) A repository of segmented and reconstructed 3D models of anatomical structures.
Discussion
References

Overview

Segmentation and reconstruction of anatomical structures from the Visible Human Datasets has generated repositories of 3D models of human anatomy and movies of these models for multiple purposes. These repositories need to be organized within a framework that allows browsing and retrieval of the 3D anatomical images.

An example of a system that accesses images organized within such a framework is the Vesalius Anatomy Browser (VAB) ([3], [5], [6]), a web-based tool that displays an image of an anatomical structure reconstructed from the Visible Human dataset in one frame and a conceptual representation of how the structure is related to other structures in another frame. The VAB accesses two types of data:

1) A navigational ontology

In our knowledge base, anatomical information is represented in an ontology, a formalism for representing commonsense or domain knowledge that supports human-like 'reasoning' by a computer system. If a user asks, for example, to see the component structures of the urinary system, the system accesses the ontology in order to determine the set of anatomical structures that satisfy the user's query. The Vesalius Navigational Ontology includes the University of Washington Digital Anatomist data, downloaded from the UMLS, and merged with the locally developed Wacholder Venuti ontology [5].

2) A repository of segmented and reconstructed 3D models of anatomical structures.

The system obtains from the navigational ontology a list of anatomical structures. In order to display arbitrary substructures in correct spatial relation to other structures, we are creating a repository of models specified in terms of the slices and binary masks used for reconstruction of 3D models from the Visible Human dataset. These reconstructed models can be combined to form 3D scenes of arbitrary complexity. [1].

To minimize storage requirements and segmentation demands, it is preferable to reduce the number of segmented elements, i.e., the basic structures that are not further decomposable in our system. We identify two properties that distinguish certain anatomical structures: decomposable and subdividable. Decomposable structures are separated by boundaries that are visible to the naked eye or can be discovered on close physical examination. For example, the pelvic viscera can be decomposed into the bladder, urethra and ureters. Each structure into which a larger structure can be decomposed must be segmented: these are the fundamental building blocks. Subdividable structures are an integral part of the larger structure in that they cannot 'naturally' be separated from the larger whole, either because the boundary between the subdivisions is conceptual (as with the penile, membranous and prostatic urethra) or because the subdivision cannot naturally be separated from the larger whole of which it is part, as with the trigone, a smooth area of the bladder's epithelial lining.

Discussion

The complexity of the problem of representing part-whole relations for concrete anatomical structures is illustrated in Figure 1, Figure 2 and Figure 3, where the ellipse S represents a 3D anatomical structure. For example, S might be taken to represent the pelvis, a relatively small though complex region of the body for which we have identified about 600 substructures covered in the unit on the pelvis for first year medical students (April 1997). Figure 1 represents a decomposable structure that we call S.

S can be subdivided into four independent structures, A, B, C and D, as indicated by the different textures, and can be recombined only in the configuration shown in Fig. 1. The lines separating the components represent borders which are either visible to the naked eye or can be discovered on close physical examination. Just as S can be rotated 360 degrees along any axis, so can A, B, C and D.

In contrast, subdivisions are integral parts of the larger structure either because they are made of the same type of tissue as the larger structure of which they are part, or are inseparable for some other reason. Their significance is associated primarily with their placement relative to the larger structure. Figure 2 illustrates one type of subdivision, where the boundaries between the subdivisions are determined conceptually, rather than on the basis of visible boundaries.

For example, the urethra is decomposed into three subdivisions: penile, prostatic and membranous. The boundaries of these subdivisions are determined by adjacent anatomical structures. Figure 3 illustrates another type of subdivision, which we call a landmark or marker, where a subdivision, though visible, has no distinct structural boundary and cannot be sensibly visualized separately from the larger whole.

Examples of landmarks are the trigone, a smooth area of the bladder's epithelial lining, and the ischial spine, a part of the pelvic bones.

Due to these distinct properties, decomposable and sub-dividable structures must be treated differently in a system in which images of anatomical structures can be taken apart and reconstructed. Decomposable structures must be segmented and visualized as distinct units; subdividable structures are visualized as highlighted regions on the surface of the object.

In the longer version of this paper, we describe in more detail how these distinctions are stored in the Vesalius Navigational Ontology, and how they drive the presentation to the user of models derived from the Visible Human dataset.