基于云服务的临床文档结构化系统设计与实现

发布时间：2019-03-08 20:18

【摘要】：随着医院信息化建设的推进,各医院都积累了大量的临床文档,已成为医疗大数据的重要组成部分。对临床文档的分析和挖掘,在疾病诊断、治疗以及病情预测上具有实际意义。由于临床文档本质上是一种非结构化文本数据,在分析与挖掘之前,往往需要通过临床文档结构化处理系统抽取出其中的结构化信息,形成可供数据分析工具和机器学习算法使用的规范化数据。众多医院在为患者提供医疗服务的同时积累了大量的临床文档,如何为这些医院提供高效的临床文档结构化服务是急需解决的问题。本课题的前期工作已对集中式临床文档结构化处理进行了研究,也提出了单机版本的临床文档结构化处理算法,构建了集中式的临床文档结构化系统。但是,以单机的方式为众多医院提供海量临床文档结构化处理服务,不管在处理速度还是存储容量方面都无法满足用户的需求。因此,本文在前期工作的基础上,构建了基于云服务的分布式临床文档结构化系统,解决了目前临床文档结构化处理过程中存在的处理速度慢,存储容量小以及数据难共享的问题。该系统为不同医院提供分布式临床文档存储服务和临床文档分布式结构化处理服务。具体而言,该系统采用了Hadoop分布式架构和OpenStack云计算管理平台,其中Hadoop架构提供了分布式临床文档存储服务和结构化处理服务,OpenStack提供了计算资源的合理分配以及临床文档的共享服务。本文的主要工作包括:(1)介绍了非结构化临床文档,阐述了超声类型临床文档的组成结构,并给出了超声文档结构化处理思路。在此基础上,提出了基于云服务的临床文档结构化系统整体框架。该系统由五个模块组成,分别为云资源管理模块、非结构化临床文档上传模块、临床文档存储及管理模块、临床文档结构化模块和结构化数据存储及管理模块,本文分别阐述了以上五个模块的设计思想和实现流程。(2)详细阐述了临床文档结构化模块的设计与实现过程。该过程主要由两个阶段组成,第一阶段对不同样本的临床文档进行指标模板的提取,第二阶段实现基于MapReduce的临床文档分布式结构化处理。其中,Map阶段包括对数据的预处理、短句切分、指标模板套用和Key,Value值的确定四个步骤,实现了对每条临床文档数据的结构化处理。Reduce阶段对分布式结构化处理后的数据按照Key值进行汇总。(3)阐述了基于云服务的临床文档结构化系统的部署与实现。给出了OpenStack结合Hadoop的具体部署过程,阐述了本系统中各个模块的具体实现过程。接着,展示了供用户操作的Web界面及其实现方式,最后,对本系统的运行效果进行评估。
[Abstract]:With the advance of hospital information construction, various hospitals have accumulated a large number of clinical documents, which has become an important part of medical big data. The analysis and mining of clinical documents are of practical significance in the diagnosis, treatment and prediction of the disease. Because the clinical document is essentially a kind of unstructured text data, before analyzing and mining, it is often necessary to extract the structured information from the structured processing system of the clinical document. Form standardized data that can be used by data analysis tools and machine learning algorithms. Many hospitals have accumulated a large number of clinical documents while providing medical services to patients. How to provide efficient structured services for clinical documents in these hospitals is an urgent problem to be solved. The previous work of this paper has carried on the research to the centralized clinical document structured processing, also has proposed the single machine version clinical document structured processing algorithm, has constructed the centralized clinical document structured system. However, a large amount of structured processing services for clinical documents are provided to many hospitals in a stand-alone manner, which can not meet the needs of users in terms of processing speed or storage capacity. Therefore, on the basis of the previous work, this paper constructs a distributed clinical document structured system based on cloud services, which solves the slow processing speed existing in the structured processing of clinical documents at present. Small storage capacity and data sharing problems. The system provides distributed clinical document storage service and clinical document distributed structured processing service for different hospitals. Specifically, the system adopts Hadoop distributed architecture and OpenStack cloud computing management platform, in which Hadoop architecture provides distributed clinical document storage services and structured processing services. OpenStack provides reasonable allocation of computing resources and sharing of clinical documents. The main work of this paper is as follows: (1) the unstructured clinical documents are introduced, the composition and structure of ultrasonic clinical documents are expounded, and the idea of structural processing of ultrasonic documents is given. On this basis, the whole framework of clinical document structured system based on cloud service is proposed. The system consists of five modules: cloud resource management module, unstructured clinical document uploading module, clinical document storage and management module, clinical document structured module and structured data storage and management module. In this paper, the design idea and implementation flow of the above five modules are described respectively. (2) the design and implementation process of the structured module of clinical documents is described in detail. The process mainly consists of two phases. In the first stage, the index templates are extracted from the clinical documents of different samples, and the distributed structured processing of clinical documents based on MapReduce is implemented in the second stage. Among them, the Map stage includes four steps: data preprocessing, short sentence segmentation, indicator template application and Key, value determination, which includes four steps: data pre-processing, short sentence segmentation, indicator template application and Key, value determination. In the reduce stage, the distributed structured data are summarized according to the key values. (3) the deployment and implementation of a structured clinical document system based on cloud services is described. The concrete deployment process of OpenStack combined with Hadoop is given, and the realization process of each module in the system is described. Then, the Web interface for user operation and its implementation are shown. Finally, the operating effect of the system is evaluated.
【学位授予单位】：东华大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP311.52;TP393.09

【参考文献】