形態素解析における放射線技術学分野の用語適用: 診療放射線技師試験を対象とした未知語の調査

摘要

Although large medical texts are stored in electronic format, they are seldom reused because of the difficulty of processing narrative texts by computer. Morphological analysis is a key technology for extracting medical terms correctly and automatically. This process parses a sentence into its smallest unit, the morpheme. Phrases consisting of two or more technical terms, however, cause morphological analysis software to fail in parsing the sentence and output unprocessed terms as "unknown words." The purpose of this study was to reduce the number of unknown words in medical narrative text processing. The results of parsing the text with additional dictionaries were compared with the analysis of the number of unknown words in the national examination for radiologists. The ratio of unknown words was reduced 1.0% to 0.36% by adding terminologies of radiological technology, MeSH, and ICD-10 labels. The terminology of radiological technology was the most effective resource, being reduced by 0.62%. This result clearly showed the necessity of additional dictionary selection and trends in unknown words. The potential for this investigation is to make available a large body of clinical information that would otherwise be inaccessible for applications other than manual health care review by personnel.%近年，病院情報システムの電子化が進み，電子化された医学文書（医学で生じるすべての文書を以下医学文吉と定義する）が日々蓄積されている．放射線医学分野では，診療画像の言売影レポート等が研究対象となっており，現在までに放射線読影レポートの形態素解析や，形態素解析に基づいた用語間の意味関係の同定等の研究が行われている．