A classification manager for compositional concept systems exemplarily shown by the AO/ASIF classification of fractures of long bones.
- 1 January 1995
- journal article
- p. 1655
Abstract
Conventional classification and coding systems represent concept systems by strict hierarchical enumeration and are supported by meaningful codes. Compositional classification is a means for representing concept systems by semantic descriptions. Classification is based on the structure of concept descriptions and explicit hierarchical relationships between their constituents. A classification manager will be presented which is based on the BERNWARD model [1]. BERNWARD is a conceptual graphs formalism and allows the constrained composition of concept descriptions by primitive concepts and roles. It stresses the distinction between generic and partitive relations. Concept descriptions can be classified on the basis of structural criteria for subsumption and part-whole relation. The capabilities of the model, compared to the principles of conventional classification and coding systems, will be exemplified by the AO/ASIF classification of fractures of long bones [2]. This classification is based on 2 axes: topography (long bone and segment) and morphology (type, group, subgroup and quality). It consists of the enumeration of all relevant fractures of long bones which are represented by a compositional meaningful code and by a line drawing. In the demonstrated system, the composition of fracture descriptions is supported by lists of terms and by graphics. The interactive selection of concepts from the space of concepts defined by the implemented classifications is supported by combining the following strategies: entering terms, selecting graphics, adding relevant characteristics to a concept selected before, and navigating through various hierarchies e.g.,generic or partitive hierarchies. These strategies are controlled by different types of compositional restrictions which are: role restrictions, hierarchical restrictions, and coordination restrictions. Role restrictions constrain the addition of specializing characteristics to elements of concept descriptions e.g., the possible complexities of fractures are simple, wedge, and multifragmented. Hierarchical restrictions constrain the generic or partitive refinement of concept elements. For example, every long bone can have the segment "proximal metaphyseal," but only tibia/fibula can have the region "malleolar." Coordination restrictions constrain the coordination of concepts e.g., a frontal fracture of the capitellum can affect the trochlea. Therefore, it is allowed to define a frontal fracture of capitellum and trochlea. Medical observations can be documented by association of selected concepts e.g., a fracture of radius and ulna can be associated to the patient Mr. X. The tools implemented for interactive selection of concepts can be used to rescan documented cases. In contrast to common classification systems, a case can be selected by combining different criteria in BERNWARD. It is possible to look for all female patients with a complex fracture of a long bone of the left upper extremity or to look for all bifocal fractures of the forearm with a wedge fracture of the radius. There is no problem to add new knowledge to the classification manager in the form of another classification. A new classification can be built by using elements of old ones e.g., the classification of the human skeleton is also useful for the AO/ASIF classification. Therefore, parts of classifications can be stored on and loaded from the disk by the demonstrated system. he user environment does not have to be changed to document a different area of medicine because of the conceptual representation of medical knowledge in BERNWARD. The user front-end can be used for all classifications e.g., a relational DBMS for Apple Macintosh systems. Some recursive functions are implemented in a linked Prolog system for effective computation of formal relations between concepts.This publication has 0 references indexed in Scilit: