Towards a Versatile Medical-Annotation Guideline Feasible Without Heavy Medical Knowledge: Starting From Critical Lung Diseases

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 4565–4572 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 4565 Towards a Versatile Medical-Annotation Guideline Feasible Without Heavy Medical Knowledge: Starting From Critical Lung Diseases Shuntaro Yada,1∗ Ayami Joh,1∗ Ribeka Tanaka,2∗ Fei Cheng,2 Eiji Aramaki,1 Sadao Kurohashi2 1Institute for Research Initiatives, Nara Institute of Science and Technology (NAIST) 8916-5, Takayama-cho, Ikoma, Nara 630-0192, Japan {s-yada, ajoh, aramaki}@is.naist.jp 2Kyoto University Yoshida-honmachi, Sakyo-ku, Kyoto, 606-8501, Japan {ribeka.tanaka, feicheng, kuro}@i.kyoto-u.ac.jp Abstract Applying natural language processing (NLP) to medical and clinical texts can bring important social benefits by mining valuable information from unstructured text. A popular application for that purpose is named entity recognition (NER), but the annotation policies of existing clinical corpora have not been standardized across clinical texts of different types. This paper presents an annotation guideline aimed at covering medical documents of various types such as radiography interpretation reports and medical records. Furthermore, the annotation was designed to avoid burdensome requirements related to medical knowledge, thereby enabling corpus development without medical specialists. To achieve these design features, we specifically focus on critical lung diseases to stabilize linguistic patterns in corpora. After annotating around 1100 electronic medical records following the annotation scheme, we demonstrated its feasibility using an NER task. Results suggest that our guideline is applicable to large-scale clinical NLP projects. Keywords: Electronic medical records, Medical annotation, Medical health record, Named entity recognition, Radiography in- terpretation report 1. Introduction Electronic medical records (EMR) are now replacing paper- based records in hospitals. Correspondingly, natural lan- guage processing (NLP) techniques in the medical field have radically gained in importance. Current NLP in med- ical fields, however, particularly addresses fundamental tasks, such as named entity (NE) recognition, NE normal- ization (also known as ontology linking), and event factu- ality detection. These tasks have been investigated indi- vidually, thereby yielding less-standardized task definitions across different studies. Table 1 portrays a summary of ex- isting corpora. As the table shows, the corpora are designed for an idiosyncratic task. Only task-specific annotation was done for each task. For example, the i2b2 Deid corpus in- cludes only personal health information (PHI). The definition (range) of tags also presents numerous varia- tions. In medical NE recognition (NER) tasks, for instance, disease names are a popular NE category. The definition of ‘diseases,’ however, varies substantially along with down- stream applications. On the one hand, disease names re- fer strictly to diagnoses of patients (Wakamiya et al., 2019; Patel et al., 2018). On the other hand, disease names en- compass not only diagnoses, but also symptoms and even complaints (Morita et al., 2013; Aramaki et al., 2014). Several corpora have been designed for general purposes covering multiple categories. An example is a Japanese medical corpus (Aramaki et al., 2009) consisting of date- time information (TIMEX), diseases, medication (drug names & dosages), test names, and test values, as well as the relation between drug names and their adverse ef- fects. Because this complex corpus …

Content truncated to 500 words