Development of a three-tierd ontology framework to realize high-quality knowledge extraction

2018 Fiscal Year Final Research Report

• Project/Area Number: 15K08845
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Medical and hospital managemen
• Research Institution: National Institute of Public Health (2018); Ehime University (2015-2017)
• Principal Investigator: Kimura Eizen 国立保健医療科学院, その他部局等, 統括研究官 (20363244)
• Co-Investigator(Kenkyū-buntansha): 岡本 和也 京都大学, 医学研究科, 准教授 (60565018) ; 今井 健 東京大学, 大学院医学系研究科(医学部), 准教授 (90401075)
• Project Period (FY): 2015-04-01 – 2019-03-31
• Keywords: 自然文章解析 / 機械学習 / オントロジー / ターミノロジー / 意味分散表現

Outline of Final Research Achievements

Acquiring an ability that can infer the causes and symptoms leading to an adverse event is required to discover potential adverse events from medical records. We tried to develop the methods to test the hypothesis that a sentence with distributed representations of words in vectors similar to the vectors calculated from sentences defining adverse events may include adverse events. In order to obtain quality distributed representations of medical terms, we analyzed the contents of electronic medical records. We developed a method for extracting mapping candidates by clustering and detecting outlier of distributed representation of words in vectors to map between UMLS; the ontology defines inter-concept relationships and medical terms. Binding medical terms with the ontology is intended to enable inferring from peripheral concepts without direct concept by back-tracing peripheral concepts from hierarchical relationships of concepts.

Free Research Field

医療情報学

Academic Significance and Societal Importance of the Research Achievements

医療記録から潜在的な事象を読み取る能力を獲得することは人工知能が医師の判断能力に近づくために必要なプロセスである。良質な意味分散表現は医療記録の分析から得られること、周辺概念からの推察を実現するためにオントロジーとの連携が重要であること、質の良い意味分散表現の獲得にむけて、表記揺れの収束や異常値の検出等の追加的な処理を加えることの必要性の知見を本研究は示した。