Extensions of SNOMED taxonomy abstraction networks supporting auditing and complexity analysis

The Systematized Nomenclature of Medicine – Clinical Terms (SNOMED CT) has been widely used as a standard terminology in various biomedical domains. The enhancement of the quality of SNOMED contributes to the improvement of the medical systems that it supports. In previous work, the Structural Analysis of Biomedical Ontologies Center (SABOC) team has defined the partial-area taxonomy, a hierarchical abstraction network consisting of units called partial-areas. Each partial-area comprises a set of SNOMED concepts exhibiting a particular relationship structure and being distinguished by a unique root concept. In this dissertation, some extensions and applications of the taxonomy framework are considered. Some concepts appearing in multiple partial-areas have been designated as complex due to the fact that they constitute a tangled portion of a hierarchy and can be obstacles to users trying to gain an understanding of the hierarchy’s content. A methodology for partitioning the entire collection of these so-called overlapping complex concepts into singly-rooted groups was presented. A novel auditing methodology based on an enhanced abstraction network is described. In addition, the existing abstraction network relies heavily on the structure of the outgoing relationships of the concepts. But some of SNOMED hierarchies (or subhierarchies) serve only as targets of relationships, with few or no outgoing relationships of their own. This situation impedes the applicability of the abstraction network. To deal with this problem, a variation of the above abstraction network, called the converse abstraction network (CAN) is defined and derived automatically from a given SNOMED hierarchy. An auditing methodology based on the CAN is formulated. Furthermore, a preliminary study of the complementary use of the abstraction network in description logic (DL) for quality assurance purposes pertaining to SNOMED is presented. Two complexity measures, a structural complexity measure and a hierarchical complexity measure, based on the abstraction network are introduced to quantify the complexity of a SNOMED hierarchy. An extension of the two measures is also utilized specifically to track the complexity of the versions of the SNOMED hierarchies before and after a sequence of auditing processes