Development of searching similar cases from clinical database with relational infomation of medical terms

目的：开发1)自动提取计算机入院报告中包含的重要临床术语。2)通过保持语义关系网络来表征病例的临床意义。在提取的医学术语中，3)从数据库中提取相似病例，4)使用交叉验证法对上述方法的效率进行评估。方法：分析提取的医学术语的语义结构，构建从临床、报告和病历中提取的医学术语的同义词库，为开发基于计算机的入院报告系统做准备。此外，我们开发了标准化的疾病临床术语，并构建了术语的结构化层次结构。结果：构建了一个包含约20000个疾病名称和60000个索引词的大型关系语义结构。我们使用这个数据库分析了800份入院报告，并提取了重要的医学术语和共现图。采用验证法对结果进行评价，评价结果为良17例，差3例。讨论：本研究的主要贡献在于构建了医学术语的语义库，为今后医学文本库的自然语言处理研究奠定了基础。我们必须加上体征和症状的术语。对于相似病例的自动提取，是否存在体征或症状是最重要的信息，而为此目的构建必要的数据库是未来需要实现的问题

Purpose : development of 1)automatic extraction of significant clinical terms contained in computer-based admission report of patients. 2)representation of clinical meaning of cases with keeping semantic network of relationships. among medical terms extracted, 3)extraction of similar cases from the database, 4)evaluation of the efficiency of the above methods using cross-validation methods.Methods : For the preparation of developing a compute-based admission reporting system, we analyzed the semantic structure of the extracted medical terms, and constructed a thesaurus database of medical terms extracted from clinical, report and medical records. Further, we developed a standardized clinical terminology of diseases and constructed the structured hierarchy of the terminology. Finally we developed database of an admission reports for the evaluation study and extracted the relations of medical terms.Results : We constructed a large relational semantic structure that contains about 20000 disease names and 60000 indexing terms. We analyzed 800 admission reports using this database and extracted significant medical terms and co-occurrence graph. Using validation method, we evaluated the results and 17 cases were good and 3 cases were evaluated as poor cases.Discussion: The main contribution of this study was that we constructed the semantic database of medical terminology for the future study for natural language processing of medical text-base. We have to add terminology of signs and symptoms. Concerning automatic extraction of similar cases, whether a sign or a symptom exists or not is the most significant information and constructing the necessary database for the purpose is the future issue to be achieved