earticle

영어

Clinical Document Processing is growing importance because of unstructured nature of clinical notes as well as limitation of crucial time of clinical professionals to analyses the unstructured clinical notes. Named entity recognition (NER) is a subtask of Clinical documentation processing which is important not only for text analysis but knowledge extraction. Although there are a number of clinical named entity recognition systems, they lack user flexibility and NER scalability. Clinical NER is a challenging work which required consistent research to improve clinical documentation. Accordingly, in this paper, keeping an eye on user’s flexibility, we combined the NER technique with DSL (Domain Specific Language) based user queries. This research focused to produce a prototype system which allows the user to input their queries about a clinical text in a syntax free language which will be reformulate into DSL format in background. The reformulated query then matches against the rules defined by using the DSL to get the matched rule-type. The DSL is created using Xtext framework specifically to create NER rules easily. Then NER is done as per the found NER rule-types. We used the lingpipe API to do the NER using unsupervised technique (dictionary based approach). Again considering user flexibility, research also focused on graphical visualization of the annotated recognized entities, flexibility to store the annotated document into database for later use as well as can conversion the recognized entities into CDA (Clinical Document Architecture) format for interoperability. This research is initiated and inspired by the Aurora research initiative which is an ongoing attempt lead by Dr. Arnold Kim to integrate the design of clinical documentation workflows from the physician perspective that starts with variety DSLs and ends with series of interpretations and analytics in the background