DEVELOPMENT OF A SEMANTIC PARSER FOR MEDICAL TEXT

该项目旨在开发一个自然语言理解系统 专门针对从医疗机构提取相关临床事实 报告。该系统主要基于语义解析技术，该技术 强调以四种形式编码的医学知识的使用。这些表格 包括：嵌入在通用医疗数据中术语的层次结构 词典；一个语义网络，旨在获取关于 不同解剖部位的相对位置；框架的集合 指定允许的术语组合。这些帧还具有 分层组织，旨在帮助解析器找到合适的 用于识别和存储复杂医学事实的格式；a 附加到框架层次的转换语法，它可以 提出医学事实的不同方式，如组合所示 一个框架中的术语，可能会被表达出来；形成了一个因果网络 特别是为了消除许多不完全表达的歧义 可以在医学报告中找到的事实。 表示系统已知的不同单词的词典和 表达报告中预期的所有有意义短语的同义词词典 域也将被建立。 使用这些信息来解析医学文本的系统将是 构造和评估。测试的域名将是CHAK的报告 慢性阻塞性肺疾病患者的X线检查、入院病史和体格检查 肺部和/或心脏疾病。评估将决定是否 报告中提供的相关医学事实由以下人员捕获和存储 集成的通用医疗数据中的自然语言解析器 基地。这个项目的目标是进一步的技术，使 将以自由文本形式捕获的医疗信息编码为表单 适用于研究、质量、保证和直接临床决策 支持。

This project seeks to develop a natural language understanding system specifically aimed at extracting relevant clinical facts from medical reports. The system is based largely on a semantic parsing technique that stresses the use of medical knowledge encoded in four forms. These forms include: a hierarchy of terms embedded in a general purpose medical data dictionary; a semantic network designed to capture knowledge concerning the relative locations of different anatomic sites; a collection of frames specifying allowable combinations of terms. These frames also have a hierarchial organization designed to help the parser find an appropriate format for the recognition and storage of a complex medical fact; a transformational grammar attached to the hierarchy of frames which can propose the different ways a medical fact, as indicated by the combined terms in a frame, might be expressed; a causal network developed specifically to allow disambiguation of the many incompletely expressed facts that can be found in a medical report. Both a lexicon expressing the different words known to the system and a thesaurus expressing all meaningful phrases expected in the reporting domain will also be built. A system that uses this information to parse medical text will be constructed and evaluated. The domains tested will be the reports of chest x-rays and admitting history and physical examination for patients with pulmonary and/or cardiac diseases. The evaluation will determine whether relevant medical facts presented in the reports are captured and stored by the natural language parser in an integrated, general purpose medical data base. The goal of this project is to further techniques that allow the encoding of medical information captured as free text into a form appropriate for research, quality, assurance, and direct clinical decision support.