SZTE-NLP: Clinical Text Analysis with Named Entity Recognition

Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), pages 615–618, Dublin, Ireland, August 23-24, 2014. SZTE-NLP: Clinical Text Analysis with Named Entity Recognition Melinda Katona and Rich´ard Farkas Department of Informatics University of Szeged ´Arp´ad t´er 2., Szeged, 6720, Hungary {mkatona,rfarkas}@inf.u-szeged.hu Abstract This paper introduces our contribution to the SemEval-2014 Task 7 on ”Analysis of Clinical Text”. We implemented a sys- tem which combines MetaMap taggings and Illinois NER Tagger. MetaMap is de- veloped to link the text of medical doc- uments to the knowledge embedded in UMLS Metathesaurus. The UMLS con- tains a very rich lexicon while the promise of a NER system is to carry out context- sensitive tagging. Our system’s perfor- mance was 0.345 F-measure in terms of strict evaluation and 0.551 F-measure in terms of relaxed evaluation. 1 Introduction Clinical notes and discharge summaries from the patient’s medical history contain a huge amount of useful information for medical researchers and also for hospitals. The automatic identification of these unstructured information is an impor- tant task for analysis of free-text electronic health records. Natural Language Processing (NLP) techniques provide a solution to process clinical documents and to help patients understand the contents of their clinical records (Tang et al., 2012; Lee et al., 2004). In this paper we introduce an approach which discovers mentions of disorders in the free-text of discharge summaries. The system participated in the SemEval-2014 Task 7: Analysis of Clinical Text, Task A. Task A aims at the identifying of mention concepts that belong to the UMLS (Boden- reider, 2004) semantic group ”disorders” and Task B is for mapping from each mention to This work is licensed under a Creative Commons At- tribution 4.0 International Licence. Page numbers and pro- ceedings footer are added by the organisers. Licence details: http://creativecommons.org/licenses/by/4.0/ a unique UMLS/SNOMED-CT CUI (Concept Unique Identifiers). Here are a few examples from the task description: • The rhythm appears to be atrial fibrillation. ,,atrial fibrillation” is a mention of type disor- ders with CUI C0004238 • The left atrium is moderately dilated. ,,left atrium [...] dilated” is a mention of type disorders with CUI C0344720 • 53 year old man s/p fall from ladder. ,,fall from ladder” is a mention of type disor- ders with CUI C0337212 Many approaches have been published to solve these problems cf. (Skeppstedt et al., 2012; Pes- tian et al., 2007). 2 Approach After a text-normalization step we run a Named Entity Recogniser (NER) on the documents. This NER model was trained on the training set of the shared task. It also employs a dictionary gathered from UMLS through MetaMap tagging. Our ini- tial experiments revealed that MetaMap (Aronson and Lang, 2010) in its own gives a very poor pre- cision hence we decided to investigate a NER ap- proach which takes the context also into account. 2.1 Normalization Clinical reports contain numerous special annota- tions, such as anonymized data (for example pa- tient name), etc. We made the following steps to normalize texts: • …

Content truncated to 500 words