The goal of this proposal is to develop an expert system for archiving medical information. The main feature of the system will be the ability to automatically index medical text with keywords. Input to the system will consist of reports from medical charts, abstracts from the medical literature, and descriptive passages from teaching collections. The system will retrieve information through requests formulated in natural language, or through Boolean combinations of keywords. The knowledge base for the expert system will be a semantic network generated from thesauri of medical terms. The expert system shell; the natural language and Boolean retrieval routines and utilities for thesaurus construction have already been developed. Partial thesauri containing an average of 3,000 terms and 15,000 links have been constructed for neuropathology, neuroradiology, and psychiatry. These thesauri have been successfully tested in pilot studies of automated indexing and natural language retrieval. During the course of this project we will complete these thesauri and compare the performance of the full autoindexing system to human indexers. If these tests are successful, we will extend thesaurus construction to neurology and neurosurgery developing a merged thesaurus covering the clinical neurosciences. The expert system will then be integrated into a comprehensive medical information system under development at the University of Pittsburgh.
