The purpose of this study is to develop and test a computer-based knowledge net as a means for the storage, indexing, and retrieval of medical information. Knowledge nets differ from traditional indexing schemes in that they contain medical information; that is, they index specific relationships betweeen medical entities, rather than merely the entities themselves. It is argued that the development of effective nets is necessary to support computer-aided diagnostic programs. All such programs, regardless of their peculiar diagnostic strategy, ultimately rely on relationships known or believed to exist between medical entities, for example, between a particular clinical finding and one or more diseases. The goal of the proposed research is to develop a method by which such relationships can be specifically referenced. This would provide not only a means to document existing expert systems, but more importantly, a method to update them, by identifying new information on the specific relationships they use. The first phase of the proposed work involves writing the programs necessary to enter, edit and retrieve information from a general-purpose medical knowledge net. This work is largely a revision and expansion of an operable prototype already developed in our laboratories. An important aspect of this work will be revision of the the Standard Nomenclature of Medicine (SNOMED) codes to eliminate numerous inconsistencies and improve its retrieval characteristics. A cross-index between the old and new codes will be created and the suitability of the net for maintaining translation tables among coding systems will be explored. This work is expected to take 18 months. The second phase of the project will involve testing the knowledge net concept. For a 12-month period, all articles related to canine medicine appearing in journals indexed by the National Library of Medicine would be entered into the knowledge net. At the end of that period, the retrieval characteristics of the net and MEDLINE would be compared. Canine medicine was selected because it is within the principal investigator's field of expertise and also because it is a good test system, a microcosm of the human literature; the volume is much smaller, but still spans all clinical specialties and academic disciplines. Analysis of the net's size and other characteristics at the end of this trial would provide a reasonable basis for extrapolation to the requirements and feasibility of similar systems in human medicine.
