Futami Hikaru, Yamagishi Hiromasa, Kawaguchi Osamu, Tsukamoto Nobuhiro, Fujii Hirofumi, Kasamatsu Tomotaka, Ando Yutaka, Osada Masakazu, Kubo Atsushi
System Integration Division, Toshiba Medical Systems Corporation.
Nihon Hoshasen Gijutsu Gakkai Zasshi. 2010 Sep 20;66(9):1229-36. doi: 10.6009/jjrt.66.1229.
Radiologists often spend much time for re-reading some of the past free-text radiology reports and determining interval changes in the physical findings when creating a report for long term cases. The aim of this study was to propose the method to detect semantic similar descriptions in the free-text reports using the structuring method based on text-mining technology. In a previous study, we had developed the structuring method that can semantically analyze the free-text reports and convert them into the description unit consisting of five items: finding/diagnosis, modifier, region, regional modifier, and confidence. Our developed prototype system extracted similar descriptions from the free-text reports by calculating the similarity index between description units. We confirmed similar descriptions extracted by the system applied to free-text reports of cases which had more than one chest CT examination written in actual clinical situation. As a result, it became available to identify candidates of similar descriptions from free-text reports. In some cases regarding practical use, the similar descriptions could not be identified in the sentences which used paraphrasing or where the findings had status changes. A solution of identifying similarity in these cases was necessary to improve the method. With the presented method here, it is expected that interval changes in the findings can be visualized and applied it to support diagnosis.
放射科医生在为长期病例撰写报告时，常常需要花费大量时间重新阅读一些过去的自由文本放射学报告，并确定体格检查结果的间隔变化。本研究的目的是提出一种基于文本挖掘技术的结构化方法，用于检测自由文本报告中的语义相似描述。在之前的一项研究中，我们开发了一种结构化方法，该方法可以对自由文本报告进行语义分析，并将其转换为由五个项目组成的描述单元：发现/诊断、修饰词、区域、区域修饰词和置信度。我们开发的原型系统通过计算描述单元之间的相似性指数，从自由文本报告中提取相似描述。我们确认了该系统应用于实际临床情况下有多次胸部CT检查的病例的自由文本报告中提取的相似描述。结果，从自由文本报告中识别相似描述的候选者变得可行。在一些实际应用的案例中，在使用释义的句子或发现有状态变化的句子中无法识别相似描述。为了改进该方法，需要解决这些情况下的相似性识别问题。通过这里提出的方法，预计可以可视化检查结果的间隔变化并将其应用于辅助诊断。
