HERMES - Help physicians to Extract and aRticulate Multimedia information from li

描述(申请人提供)：医生在看病人时有很多疑问。据报道，初级保健医生每10次就诊就会产生0.7到18.5个问题。出版的医学文献是帮助医生获取最新的临床信息，从而提高患者护理质量的重要资源。例如，上例中的案例研究(即对脂肪团的诊断程序和治疗)发表在《新英格兰医学杂志》(NEJM)的一篇《临床实践》文章中。尽管大型医院的医生经常使用PubMed，但它不会返回具体问题的答案。通常，PubMed会返回大量文章来响应特定的用户查询。医生浏览检索到的文章的时间有限；已经发现，医生平均花费两分钟或更少的时间来寻找问题的答案，如果搜索花费更长的时间，很可能会被放弃。一项评估研究表明，医疗保健提供者从PubMed搜索答案平均需要30分钟以上，这使得“寻找信息……只在‘下班后’才实用，而不是在临床环境中。”人们得出的结论是，缺乏时间是导致许多医学问题悬而未决的最常见障碍。 在病人护理的角度回答医生的问题的重要性已经被医学界广泛认识到。许多医学数据库(如UpToDate和Thomson Micromedex)提供摘要，以回答与患者护理相关的重要医学问题。然而，大多数摘要是由手动审查文献信息的医学专家撰写的。这些数据库在范围和及时性方面都是有限的。 我们假设我们可以开发医疗语言处理(MLP)方法来构建一个完全自动化的系统Hermes-帮助医生从文献中提取和清楚地表达多媒体信息，以回答他们特别的医疗问题。Hermes将自动检索、提取、分析和集成文献中的文本、图像和视频，并将其表述为医生提出的临时医学问题的答案。我们的初步结果表明，即使是一个有限的Hermes工作系统也比其他信息检索系统性能更好，并且可以在必要的时间框架内生成满足医生需求的答案。爱马仕承诺协助医生实践循证医学(EBM)，这是一种涉及明确使用当前最佳证据的医疗实践，即主要医学文献中报告的以患者为中心的高质量临床研究。 我们的具体目标是： 1)从即席医疗问题中确定信息需求。我们将结合丰富的语义、统计和机器学习方法，自动将即席医学问题映射到其组成部分问题类型。组成部分问题类型是一种通用的简单问题类型，它需要不同于其他组成部分问题类型的回答策略。 2)开发新的信息检索模型，该模型集成了特定领域的知识来检索相关文档，以回答特定的医学问题。 3)从检索到的文档中提取相关文本、图像和视频，以回答特定的医学问题。 4)集成文本、图像和视频，融合信息以生成简短而连贯的多媒体摘要。 5)设计可用性研究，以衡量Hermes的有效性、准确性和感知的易用性，并将Hermes与其他信息系统进行比较。

DESCRIPTION (provided by applicant): Physicians have many questions when seeing patients. Primary care physicians are reported to generate between 0.7 and 18.5 questions for every 10 patient visits. The published medical literature is an important resource helping physicians to access up-to-date clinical information and thereby to enhance the quality of patient care. For example, the case study in the above example (i.e., diagnostic procedures and treatment for cellulites) was published in a "Clinical Practice" article in the New England Journal of Medicine (NEJM). Although PubMed is frequently used by physicians in large hospitals, it does not return answers to specific questions. Frequently, PubMed returns a large number of articles in response to a specific user query. Physicians have limited time for browsing the articles retrieved; it has been found that physicians spend on average two minutes or less seeking an answer to a question, and that if a search takes longer it is likely to be abandoned. An evaluation study has shown that it takes an average of more than 30 minutes for a healthcare provider to search for answer from PubMed, which makes "information seeking ... practical only `after hours' and not in the clinical setting." It has been concluded that a lack of time is the most common obstacle resulting in many unanswered medical questions. The importance of answering physicians' questions at the point of patient care has been widely recognized by the medical community. Many medical databases (e.g., UpToDate and Thomson MICROMEDEX) provide summaries to answer important medical questions related to patient care. However, most of the summaries are written by medical experts who manually review the literature information. The databases are limited in their scope and timeliness. We hypothesize that we can develop medical language processing (MLP) approaches to build a fully automated system HERMES - Help physicians to Extract and aRticulate Multimedia information from literature to answer their ad-hoc medical quEstionS. HERMES will automatically retrieve, extract, analyze, and integrate text, image, and video from the literature and formulate them as answers to ad-hoc medical questions posed by physicians. Our preliminary results show that even a limited HERMES working system outperformed other information retrieval systems and can generate answers within a timeframe necessary to meet the demands of physicians. HERMES promise to assist physicians for practicing evidence-based medicine (EBM), the medical practice that involves the explicit use of current best evidence, i.e., high-quality patient-centered clinical research reported in the primary medical literature. Our specific aims are: 1) Identify information needs from ad-hoc medical questions. We will incorporate rich semantic, statistical, and machine learning approaches to map ad-hoc medical questions to their component question types automatically. A component question type is a generic, simple question type that requires an answer strategy that is different from other component question types. 2) Develop new information retrieval models that integrate domain-specific knowledge for retrieving relevant documents in response to an ad-hoc medical question. 3) Extract relevant text, images, and videos from the retrieved documents in response to an ad-hoc medical question. 4) Integrate text, images, and videos, fusing information to generate a short and coherent multimedia summary. 5) Design a usability study to measure efficacy, accuracy and perceived ease of use of HERMES and to compare HERMES with other information systems.