A Method of Medical Knowledge Base Augmentation

The similarity portion of the experiment showed that a method of merging based on similarity of features is a workable approach to incorporating knowledge into MeSH when lexical matches are not available.

Abstract

Summary The Medical Subject Headings (MeSH) of the National Library of Medicine may be viewed as a semantic network. The relationships in this semantic network are of a broader-than/narrower-than type. A knowledge base of this type may be augmented by adding new terms and new relationships to the network. The Current Medical Information and Terminology (CMIT) of the American Medical Association represents a rich source of relationships for the disease terms of MeSH. A subset of MeSH was augmented with the knowledge from a subset of CMIT using a matching and similarity strategy. The matching portion of the experiment showed that about half of CMIT may be directly merged with MeSH based on exact and partial matches and utilization of alternate and synonym terms from CMIT. The similarity portion of the experiment showed that a method of merging based on similarity of features is a workable approach to incorporating knowledge into MeSH when lexical matches are not available. Evaluation of the resulting merged knowledge base suggested that the etiology property of CMIT was the most highly inherited property. The augmented knowledge base was used as a basis for an automatic indexer. The indexer was less accurate after augmentation than before. One key difficulty stemmed from the way that CMIT was encoded into MeSH. More powerful encodings of CMIT into MeSH are being pursued. Building on MeSH, CMIT, and other such knowledge bases that already exist on the computer is one way to try to develop intelligent medical computer systems.