Charles Edward Kahn MD, Presenter: Nothing to Disclose

Cheng Thao, Abstract Co-Author: Nothing to Disclose

Michael Q. Warnock, Abstract Co-Author: Nothing to Disclose

Kevin C. Ehlers MS, Abstract Co-Author: Nothing to Disclose

Most radiology reports consist of unstructured ("free") text. We explored techniques to map radiology report text to concepts in a controlled vocabulary, and to use those concepts to discover relevant knowledge resources. We tested the feasibility of an automated approach to identify pertinent articles in the radiological literature based on concepts extracted from free-text radiology reports.

The clinical indications and the impression text of 20 radiological procedure reports, selected at random from among procedures performed on a single day in 2005, were obtained from a radiology information system. No personally identifiable information was included. The MetaMap Transfer (MMTx) program, provided by the U.S. National Library of Medicine, mapped the report text to concepts in the Unified Medical Language System (UMLS) Metathesaurus, version 2003AA. We used PubMed's Entrez Programming Utilities to search the biomedical literature using these concepts. The search was limited to articles that were published in the journal Radiology or RadioGraphics from January 2000 through December 2004, had abstracts, and were indexed as being related to human subjects.

The MMTx software identified 2 to 28 UMLS Metathesaurus terms from each report; the number of concepts extracted varied in part with the length of the text block analyzed. The concept mapping algorithm required no more than 30 seconds. The Entrez Programming Utilities successfully identified articles within the journals of interest that were indexed by concepts related to those extracted from the radiology reports; the search typically took no more than 5 seconds.

We have demonstrated a rapid, automated process for discovery of pertinent articles from the radiological literature based on free-text radiology reports. By constraining the search space of journal articles, we were better able to discover materials pertinent to the radiological examination and its clinical indications. This work can be used to guide discovery of knowledge resources -- in the biomedical literature or other appropriately indexed collections -- from unstructured radiology report text.

Kahn, C, Thao, C, Warnock, M, Ehlers, K, Automated Resource Discovery from Radiology Report Text.  Radiological Society of North America 2005 Scientific Assembly and Annual Meeting, November 27 - December 2, 2005 ,Chicago IL. http://archive.rsna.org/2005/4413332.html