JaMIE: A Pipeline Japanese Medical Information Extraction System with Novel Relation Annotation

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 3724–3731 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 3724 JaMIE: A Pipeline Japanese Medical Information Extraction System with Novel Relation Annotation Fei Cheng1∗, Shuntaro Yada2∗, Ribeka Tanaka3, Eiji Aramaki2, Sadao Kurohashi1 1Kyoto University, Kyoto, Japan, 2Nara Institute of Science and Technology, Nara, Japan, 3Ochanomizu University, Tokyo, Japan {feicheng, kuro}@i.kyoto-u.ac.jp, {s-yada,aramaki}@is.naist.jp, tanaka.ribeka@is.ocha.ac.jp Abstract In the field of Japanese medical information extraction, few analyzing tools are available and relation extraction is still an under-explored topic. In this paper, we first propose a novel relation annotation schema for investigating the medical and temporal relations between medical entities in Japanese medical reports. We experiment with the practical annotation scenarios by separately annotating two different types of reports. We design a pipeline system with three components for recognizing medical entities, classifying entity modalities, and extracting relations. The empirical results show accurate analyzing performance and suggest the satisfactory annotation quality, the superiority of the latest contextual embedding models. and the feasible annotation strategy for high-accuracy demand. Keywords: Medical Information Extraction, Corpus Annotation, Relation Extraction, Open-access Toolkit 1. Introduction Electronic medical record systems have been widely adopted in the hospitals. In the past decade, research efforts have been devoted to automated Information Extraction (IE) from raw medical reports. This ap- proach should be able to liberate users from the burden of reading and understanding large volumes of records manually. While substantial progress has been made already in medical IE, it still suffers from the following limitations. First, languages are the natural boundaries to hinder the existing research from being reused across lan- guages. The development of the English corpora and approaches can less reflect the progress in other lan- guages. Morita et al. (2013; Aramaki et al. (2014; Ara- maki et al. (2016) present a series of Japanese clin- ical IE shared tasks. However, more semantic-aware tasks such as medical relation extraction (Uzuner et al., 2011) and temporal relation extraction (Bethard et al., 2017) are still undeveloped. Second, most exist- ing medical IE datasets focus on general report content such as discharge summary, instead of more specific re- port types and diseases. Such settings potentially sac- rifice the accuracy for analyzing specific report types, such as radiography interpretation reports. In this work, we first propose a novel relation anno- tation scheme for investigating the medical and tem- poral relations in Japanese medical reports. Then, we intend to explore the correlation between the annota- tion efforts on specific report types and their analyzing accuracy, which is especially in demand for practical medical applications. Therefore, we target the compar- ison of analyzing two report types involved with the *These authors contributed equally to this work diseases of high death rates: (1) specific radiography interpretation reports of lung cancer (LC), (2) medi- cal history reports (containing multiple types of reports relevant to a patient) of idiopathic pulmonary fibrosis (IPF). The relation annotation is based on the existing entities presented by Yada …

Content truncated to 500 words