Daniel L. Rubin MD, Abstract Co-Author: Nothing to Disclose

Jonathan R Nebeker MD, Abstract Co-Author: Nothing to Disclose

Scott L. Du Vall PhD, Presenter: Partner, Scottwares Consulting, Salt Lake City, UT Partner, Benefits Delivered, LLC, Salt Lake City, UT

By formatting RadLex to take advantage of an existing open-source information extraction tool, and only with a few modifications to the existing tool for optimization, we were able to extract RadLex concepts from radiology reports along with clinical concepts described in UMLS.

Most interpretation of imaging studies is recorded in narrative format – either typed directly by the radiologist or transcribed from speech recognition. Information extraction (IE) is a set of techniques used to identify clinical terms in these documents. IE allows clinicians to continue to record pertinent information in a way that is natural (narrative text) and creates a formal representation of the information that computers can manipulate. IE tools have been developed both in medicine and in the natural language processing community at large to map text to standard terminologies such as those in the National Library of Medicine Unified Medical Language System (UMLS). One such system is Health Information and Text Extraction (HITEx). It is an open-source application that contains modules for processing text and identifying clinical concepts. The HITEx UMLS Concept Finder is a module that assigns UMLS codes to text by checking for matching terms, synonyms, and root words.

RadLex is not contained in the UMLS and is not fully mapped to any of its source terminologies. We formatted the RadLex terminology to match the UMLS file structure and inserted terms as custom concepts in order to take advantage of the HITEx UMLS Concept Finder module. We also made modifications to the module in order to optimize RadLex term discovery. This project discusses the system modification. Evaluation of the effectiveness of these changes in identifying RadLex concepts in radiology reports will be done in further studies.

RadLex is the lexicon developed to unify and supplement the many existing standards and terminologies used in radiology. Because RadLex fills the gap for concepts common only in imaging, information extraction on radiology reports using any other code set would miss certain medical devices, image acquisition techniques, and some fine anatomy.

Rubin, D, Nebeker, J, Du Vall, S, Modifying an Open-Source Information Extraction Tool to Identify RadLex Terms in Clinical Notes.  Radiological Society of North America 2009 Scientific Assembly and Annual Meeting, November 29 - December 4, 2009 ,Chicago IL. http://archive.rsna.org/2009/8006984.html