ULisboa: Recognition and Normalization of Medical Concepts

Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 406–411, Denver, Colorado, June 4-5, 2015. c⃝2015 Association for Computational Linguistics ULisboa: Recognition and Normalization of Medical Concepts Andr´e Leal+, Bruno Martins∗, and Francisco M. Couto+ +LASIGE, Faculdade de Ciˆencias, Universidade de Lisboa, 1749-016 Lisboa, Portugal. ∗INESC-ID, Instituto Superior T´ecnico, Universidade de Lisboa, Portugal aleal@lasige.di.fc.ul.pt , bruno.g.martins@ist.ul.pt , fcouto@di.fc.ul.pt Abstract This paper describes a system developed for the disorder identification subtask within task 14 of SemEval 2015. The developed sys- tem is based on a chain of two modules, one for recognition and another for normaliza- tion. The recognition module is based on an adapted version of the Stanford NER system to train CRF models in order to recognize dis- order mentions. CRF models were build based on a novel encoding of entity spans as token classifications to also consider non-continuous entities, along with a rich set of features based on (i) domain lexicons and (ii) Brown clus- ters inferred from a large collection of clinical texts. For disorder normalization, we (i) gen- erated a non ambiguous dictionary of abbrevi- ations from the labelled files, using it together with (ii) an heuristic method based on similar- ity search and (iii) a comparison method based on the information content of each disorder. The system achieved an F-measure of 0.740 (the second best), with a precision of 0.779, a recall of 0.705. 1 Introduction Clinical notes are an important source of informa- tion recorded by medical professionals. However, this information, when available, is not easily acces- sible within automated procedures. Clinical notes are inherently complex, due to their lack of struc- ture (i.e., narrative language) and due to the need for contextual interpretation. To address this complex- ity, text mining approaches represent an effective so- lution to assist the users in retrieving and extracting the required information. This paper presents a text mining system for pro- cessing clinical text, that we developed for SemEval based on a pipeline with two modules, one for entity recognition and another for normalization. The entity recognition module is based on the Stanford NER tool (Finkel et al., 2005), and it uses CRF models trained on annotated biomedical notes. The module tags the text according to an SBIEON encoding of entities as token classes, supporting the recognition of non-continuous entities (Leal et al., 2014). We relied on features based on Brown clus- ters and domain specific lexicons. Thus, this ap- proach combines both supervised (Stanford NER) and unsupervised methods (Brown Clusters). For practical applications, entity recognition is incomplete without performing normalization, i.e. without mapping each entity to an identifier (CUI) in a controlled vocabulary like SNOMED CT (Cornet and Keizer, 2008), that defines its semantic meaning. One of the main challenges in this task consists in re- solving the ambiguous cases, where the same entity can have distinct semantic meanings (i.e., mapped to distinct CUIs) depending on the context. Our normalization module relies on the follow- ing components: (i) a procedure for the automatic generation …

Content truncated to 500 words