Multi-source clinical Question Answering system

描述(由申请人提供)： /摘要(限制：1页)我们的建议解决了以下挑战领域：06-LM-101*用于回答临床问题的智能搜索工具。开发新的信息检索计算方法，允许临床医生或临床研究人员提出单个查询，导致搜索多个数据源，以产生一致的响应，突出关键的相关信息，这些信息可能标志着临床研究或患者护理的新见解。可以帮助临床医生诊断或管理健康状况或帮助临床研究人员探索临床试验期间出现的问题的重要性的信息分散在许多不同类型的资源中，例如纸质或电子图表、试验方案、发表的生物医学文章或护理最佳实践指南。开发人工智能和信息检索方法，使面临复杂患者问题的临床医生或研究人员能够提出一个单一的查询，该查询将导致似乎“理解”该问题的搜索，即检查多个数据库并将发现汇集在一起形成有用的答案的搜索。临床问答(CQA)系统关注的是医生的需求，通常是在护理点，或者实验室的研究人员。通常被问的问题要么需要高度特定于患者的信息，例如患者的实验室结果或以前的病史，由患者的健康记录回答，要么需要通常通过普遍可用的信息来源回答的更一般类型的信息。问答系统通过提供相关信息的简明摘要以及源命中来增强搜索引擎的结果。PubMed(http://www.ncbi.nlm.nih.gov/pubmed/)是最普遍的生物医学搜索引擎，然而，因为它是一个搜索引擎，所以检索到的信息是基于关键字搜索的，而不是以立即消费的形式呈现的；用户必须深入网页的内容以找到感兴趣的事实/陈述。此外，临床医生需要的信息可能是不同类型的，例如，结合针对特定患者的特定诊断所触发的特定行动的综合症定义。这样的信息存在于不同的来源--百科全书和电子病历--并且必须以易于理解的格式动态地访问并呈现给用户。我们建议从临床和生物医学叙事的多源开发一个统一的临床问答平台，通过融合两种现有技术-Mayo临床文本分析和知识提取系统和科罗拉多大学的问答系统来实现问题的语义处理。我们致力于回答的具体研究问题是：将一个通用的语义QA系统移植到临床领域需要付出多少努力？需要多少额外的特定领域的培训？这样一个系统的准确性是什么？临床领域的问答是一个新兴的研究领域。外地面临的挑战主要是因为需要进行特定领域培训的组件数量太多，以及在高精确度和查全率方面的严格系统要求，并辅之以便于使用和方便用户的介绍。我们克服这些问题的方法是重新使用已经存在的组件，作为Mayo临床文本分析和知识提取系统以及科罗拉多大学问题回答系统的一部分。我们的方法是创新的，将百科全书来源的信息和EMR结合在一起，以统一的形式呈现给护理点的临床医生或实验室的研究人员。这项技术是基于语义语言处理的，其目的是“理解”问题和叙述的意义。我们提议的系统有可能影响医疗保健和转化性研究的质量。我们的方法是可行的，因为它使用了梅奥诊所电子病历中已有的内容，以及来自多种现成资源的一般医学知识。拟议的系统将建立在成熟和经过测试的组件上，从而实现快速而稳健的交付周期。我们独特的技术与复杂的统计机器学习算法相结合，应用于丰富的关于事件、矛盾、语义结构和问题类型的语言学知识，将使我们能够构建一个系统，显著扩展临床医生可用的可能问题类型和回答的范围，并将这些无缝融合以生成回答。我们提议的工作代表了一个有可能改善医疗保健提供的高影响领域，因为它解决了已经被很好地记录和研究的需求(Ely等人，2005年)。我们的目标是为关注点或实验室的相关信息的语义检索、访问和摘要提供统一的多源解决方案。因此，拟议的CQA有可能对医生或生物医学研究人员起到至关重要的决策支持作用。(最多2-3句)临床问答(CQA)系统关注的是医生的需求，通常是在护理点或实验室的研究人员。通常被问的问题要么需要高度特定于患者的信息，例如患者的实验室结果或以前的病史，由患者的健康记录回答，要么需要通常通过普遍可用的信息来源回答的更一般类型的信息。我们建议的工作是提供统一的多源解决方案，用于在护理点或实验室对相关信息进行语义检索、访问和汇总，这是一个有可能改善医疗保健提供的高影响领域，因为它解决了已被很好地记录和研究的需求。

DESCRIPTION (provided by applicant): / Abstract (Limit: 1 page) Our proposal addresses the following challenge area: 06-LM-101* Intelligent Search Tool for Answering Clinical Questions. Develop new computational approaches to information retrieval that would allow a clinician or clinical researcher to pose a single query that would result in search of multiple data sources to produce a coherent response that highlights key relevant information which may signal new insights for clinical research or patient care. Information that could help a clinician diagnose or manage a health condition, or help a clinical researcher explore the significance of issues that arise during a clinical trial, is scattered across many different types of resources, such as paper or electronic charts, trial protocols, published biomedical articles, or best-practice guidelines for care. Develop artificial intelligence and information retrieval approaches that allow a clinician or researcher confronting complex patient problems to pose a single query that will result in a search that appears to "understand" the question, a search that inspects multiple databases and brings findings together into a useful answer. Clinical question answering (cQA) systems focus on the physician needs usually at the point of care, or the investigator in the lab. The questions usually asked either require information highly specific to their patient, e.g. the patient's lab results or previous history, answered by the patient's health record, or a more general type of information usually answered through generally available information sources. QA systems enhance the results of search engines by providing a concise summary of relevant information along with source hits. PubMed (http://www.ncbi.nlm.nih.gov/pubmed/) is the most ubiquitous biomedical search engine, however because it is a search engine the information retrieved is based on keyword searches and is not presented in a form for immediate consumption; the user has to drill down into the content of the webpages to find the facts/statements of interest. Moreover, the information that the clinician needs is likely to be of different types, for example a definition of a syndrome in combination with specific actions triggered by a particular diagnosis for a particular patient. Such information resides in different sources - encyclopedic and the EMR - and has to be dynamically accessed and presented to the user in an easily digestible format. We propose to develop a unified platform for clinical QA from multiple sources of clinical and biomedical narrative that implements semantic processing of the questions by fusing two existing technologies - the Mayo clinical Text Analysis and Knowledge Extraction System and the University of Colorado's Question Answering System. The specific research questions we are aiming to answer are: "How much effort is required to port a general semantic QA system to the clinical domain? How much additional domain-specific training is required? "What is the accuracy of such a system? Question Answering in the clinical domain is an emerging area of research. The challenges in the field are mainly attributed to the number of components that require domain specific training along with strict system requirements in terms of high precision and recall complemented by an accessible and user-friendly presentation. Our approach to overcome them is to re-use components already in place as part of Mayo clinical Text Analysis and Knowledge Extraction System and the University of Colorado's Question Answering System. Our approach is innovative in bringing together information from encyclopedic sources and the EMR to present it into a unified form to the clinician at the point of care or the investigator in the lab. The technology for that is based on semantic language processing which aims at "understanding" the meaning of the question and the narrative. Our proposed system holds the potential to impact quality of healthcare and translational research. Our approach is feasible because it uses content already in the EMR at the Mayo Clinic along with general medical knowledge from multiple readily-available resources. The proposed system will be built off mature and tested components allowing a fast and robust delivery cycle. Our unique integration of technologies together with sophisticated statistical machine learning algorithms applied to rich linguistic knowledge about events, contradictions, semantic structure, and question-types, will allow us to build a system which significantly extends the range of possible question types and responses available to clinicians, and seamlessly fuses these to generate a response. Our proposed work represents a high impact area that has the potential to improve healthcare delivery because it addresses needs that have been well-documented and studied (Ely et al., 2005). We aim to provide a unified multi-source solution for semantic retrieval, access and summarization of relevant information at the point of care or the lab. As such, the proposed cQA has the potential to play a vital and important decision- support role for the physician or the biomedical investigator. (max 2-3 sentences) Clinical question answering (cQA) systems focus on the physician needs usually at the point of care, or the investigator in the lab. The questions usually asked either require information highly specific to their patient, e.g. the patient's lab results or previous history, answered by the patient's health record, or a more general type of information usually answered through generally available information sources. Our proposed work to provide a unified multi-source solution for semantic retrieval, access and summarization of relevant information at the point of care or the lab, represents a high impact area that has the potential to improve healthcare delivery because it addresses needs that have been well-documented and studied.