Neuhaus Philipp, Doods Justin, Dugas Martin
Institute of Medical Informatics, Westfälische Wilhelms-Universität, Münster, Germany.
Stud Health Technol Inform. 2015;212:23-6.
Automatic coding of medical terms is an important, but highly complicated and laborious task.
To compare and evaluate different strategies a framework with a standardized web-interface was created. Two UMLS mapping strategies are compared to demonstrate the interface.
The framework is a Java Spring application running on a Tomcat application server. It accepts different parameters and returns results in JSON format. To demonstrate the framework, a list of medical data items was mapped by two different methods: similarity search in a large table of terminology codes versus search in a manually curated repository. These mappings were reviewed by a specialist.
The evaluation shows that the framework is flexible (due to standardized interfaces like HTTP and JSON), performant and reliable. Accuracy of automatically assigned codes is limited (up to 40%).
Combining different semantic mappers into a standardized Web-API is feasible. This framework can be easily enhanced due to its modular design.
医学术语的自动编码是一项重要但极其复杂且费力的任务。
为了比较和评估不同策略，创建了一个具有标准化网络界面的框架。比较了两种统一医学语言系统（UMLS）映射策略以展示该界面。
该框架是一个运行在Tomcat应用服务器上的Java Spring应用程序。它接受不同参数并以JSON格式返回结果。为了展示该框架，通过两种不同方法对一系列医学数据项进行映射：在一个大型术语代码表中进行相似性搜索与在一个人工整理的存储库中进行搜索。这些映射由一位专家进行审核。
评估表明该框架具有灵活性（由于HTTP和JSON等标准化接口）、高性能且可靠。自动分配代码的准确性有限（高达40%）。
将不同的语义映射器组合成一个标准化的网络应用程序编程接口（Web-API）是可行的。由于其模块化设计，该框架可以很容易地得到增强。
