Clinical named entity recognition: Challenges and opportunities

Information Extraction (IE), one of the important tasks in text analysis and Natural Language Processing (NLP), involves extracting meaningful pieces of knowledge from unstructured information sources, as unstructured data is computationally opaque. The intent of IE is to produce a knowledge base i.e. organize the information in a way that it is useful to people and arrange the information in a semantic way so that algorithms can make certain useful inferences from it. Named Entity Recognition (NER) is a sub-task of IE which finds and classifies the names/entities. Once these Named Entities (NE) are extracted, they can then be indexed and made searchable, relations can be derived, questions can be answered and many more. NER techniques are different for different domains, because of the uniqueness that exists in each of the domains, although the process depends on a number of fundamental Natural Language Processing (NLP) steps such as tokenization, part-of-speech tagging, parsing and model building. As an example, NER in the medical domain involves handling of a number of vital tasks such as identification of medical terms, attributes such as negation, severity, identification of relationships between entities and mapping terms in the document to concepts in domain specific ontologies. There is also a heavy dependence on domain specific resources such as medical dictionaries and ontologies such as the Unified Medical Language System (UMLS)34. In this paper, we focus on NER in the clinical domain. In particular, we will focus on the NER challenges and the qualitative analysis of clinical reports on the approaches we took for the named entities: anatomies, findings, location qualifier, and procedures.