Semantic indexing of biomedical publications and clinical text

National Library Of Medicine

Investigators

Linked publications & trials

Abstract

The Indexing Initiative project addresses both objectives by continually improving Named Entity Recognition algorithms and providing these algorithms to the public in a program called MetaMap. MetaMap is widely used both as an NLM service for remote file processing and as a downloadable tool. In 2016, the Lister Hill Center produced a new open source version of legacy MetaMap, written in JAVA, a language with a broad programmer following and immense industry support. Named MetaMapLite, it is an order of magnitude simpler and faster than its progenitor. It is also many fold faster than any other open source alternative and will be easier for others in the community to remodel and use for their own special purposes. Speed is essential for such applications as NLM InfoBot, in which MetaMapLite has replaced a tool that was using a much smaller vocabulary to get the speed at the cost of better text understanding. InfoBot is used at the Clinical Center in their EHR called CRIS to support evidence-based care plan development for interdisciplinary teams. The tool gets about 30 hits a day at any time of the day, with greater use in the off-hours when nurses have very few colleagues around to ask for their opinion. Anecdotally, InfoBot had a significant spike in use during the furlough when only essential personnel were caring for patients. In 2016, MetaMap and other related software and data were downloaded 122,636 times. In addition, the tools processed 1,044,682 documents. The semantic indexing objective is addressed by the Medical Text Indexer (MTI) program that assigns most salient medical subject headings to medical text. Among other applications, MTI powers MeSH on Demand an online tool that helps authors find key terms for their publications, identifying MeSH Terms relevant to submitted text.