One of the hypotheses underlying our research is that an ontology provides a conceptually and computationally sound  approach for linking anatomical entities and their medically significant subparts. An ontology is a type of knowledge representation system which includes common sense knowledge organized so that a computer system can perform deductive reasoning.  Loosely defined, an ontology is a set of terms, belonging to some domain of interest, associated with definitions in natural language and linked by a set of formal relations and constraints (adapted from [7]).

     In the Vesalius ontology, the terms are anatomical concepts for medically significant entities and associated images.  At present, these concepts are linked by two sorts of formal relations—definitional/classificatory (taxonomical) and part-whole (meronymical).

     The formal relationship most widely used in ontologies in all domains is generally referred to as the is-a relation, a definitional relationship between a general class (e.g., birds) and a more specific set of subclasses (e.g., canaries and penguins). In the is-a relation, each subclass shares (technically, ‘inherits') all of the properties of the more general class (its parent) and is also characterized by additional properties (differentia) that distinguish it from other subclasses (its siblings). For example, in Figure 2, birds are a class of animals that breathe with lungs and that have two wings. Canaries are a more specific subclass of birds  that inherit the general properties of birds (in this example, having feathers, two legs and wings) from the parent class.  The siblings are characterized by specific properties that distinguish them from other siblings.  In this example, the siblings, namely canaries and penguins, are differentiated by the color of their feathers and by their ability to fly.

     This classificatory relationship is important in both anatomy and computer science.  An understanding of human anatomy  depends on conceptualization of a hierarchy of physical units or parts, and of the relationships that hold between them [8].   The same relationships, formally defined in terms of inheritance properties by which classes are related , can be used to support deductive reasoning about anatomy by a computer system. To give a simple example, if a prostate is-a viscus and a viscus is-a body organ, then a computer system can be programmed to deduce that the prostate is a body organ, and the prostate can be listed whenever a user asks for a list of body organs.

     Definitional relationships like this one play a crucial role in making it possible for computers to make commonsense deductions like the one in the last paragraph.  However, they are not the only important relationship. In general, ontologies have suffered from the problem of over-use of the is-a relation, to the exclusion of others [9]. Although they are recognized as important,  part-whole relations among anatomical entities have to date received less attention in medical ontologies.  In previous work, we have identified (at least) three distinct part-whole relations needed for navigating anatomical images [10, 11].

• structured-components, where the components of the larger structure can be listed and are organized in a patterned structure.  For example, the individual bones are components of the skeletal system; these bones are arranged in a pre-determined pattern which forms the skeletal system, and each bone can be separated from the larger entity, and viewed independently.

 
• landmark, where the smaller entities are part of the mass of the larger entity and cannot be physically separated from it.  For example, the ischium has the following landmarks: spine, ramus, tuberosity, lesser notch and acetabulum.

 
• composed-of, where the smaller entity consists of the stuff that the larger entity is made of.  For example, a muscle is composed of muscle tissue and muscle tissue is composed of muscle cells.  Muscle cells can be physically separated from muscle tissue, so they are not landmarks,  and they cannot be individually listed, so they are not structured components.

     For now, we are using the is-a, structured-component and landmark relations  to test the hypothesis that an ontology provides a natural way to link anatomical entities and their medically significant subparts and therefore can serve as the basis for a system for navigating anatomical images that is useful for students.
To evaluate this hypothesis, we have developed a prototype anatomical ontology for the pelvis in which anatomical entities are connected by these relationships and linked to 3-D anatomical images reconstructed from the Visible Human Dataset.  This prototype is a ‘curriculum ontology’, an ontology developed in conjunction with a member of the medical school faculty to serve as a study-tool for a particular course or set of lectures.  A curriculum ontology includes:

• a faculty-developed list of anatomical entities covered in the classroom;
• a set of 3-D anatomical images of these entities;
• a set of formally defined relationships linking these images.