Having an entity annotated corpus of the clinical domain is one of the basic requirements for detection of clinical entities using machine learning (ML) approaches. Past researches have shown the superiority of statistical/ML approaches over the rule based approaches. But in order to take full advantage of the ML approaches, an accurately annotated corpus becomes an essential requirement. Though there are a few annotated corpora available either on a small data set, or covering a narrower domain (like cancer patients records, lab reports), annotation of a large data set representing the entire clinical domain has not been created yet. In this paper, we have described in detail the annotation guidelines, annotation process and our approaches in creating a CER (clinical entity recognition) corpus of 5,160 clinical documents from forty different clinical specialities. The clinical entities range across various types such as diseases, procedures, medications, medical devices and so on. We have classified them into eleven categories for annotation. Our annotation also reflects the relations among the group of entities that constitute larger concepts altogether.

This paper presents a treebank for the healthcare domain developed at ezDI. The treebank is created from a wide array of clinical health record documents across hospitals. The data has been de-identified and annotated for constituent syntactic structure. The treebank contains a total of 52053 sentences that have been sampled for subdomains as well as linguistic variations. The paper outlines the sampling process followed to ensure a better domain representation in the corpus, the annotation process and challenges, and corpus statistics. The Penn Treebank tagset and guidelines were largely followed, but there were many syntactic contexts that warranted adaptation of the guidelines. The treebank created was used to re-train the Berkeley parser and the Stanford parser. These parsers were also trained with the GENIA treebank for comparative quality assessment. Our treebank yielded great-er accuracy on both parsers. Berkeley parser performed better on our treebank with an average F1 measure of 91 across 5-folds. This was a significant jump from the out-of-the-box F1 score of 70 on Berkeley parser’s default grammar.