CLAMP-CS: a Cloud-based, Service-oriented, high-performance Natural Language Processing Platform for Healthcare

项目摘要 电子健康记录（EHR）的广泛采用导致了巨大的临床数据库，这使得快速 医疗分析市场的增长。分析EHR数据的一个特殊挑战是， 患者信息嵌入在临床文档中并且不能直接用于下游分析。 因此，临床自然语言处理（NLP）技术，可以解锁嵌入在 临床叙述受到了极大的关注，预计到2021年全球市场将达到26.5亿美元。在我们 在以前的工作中，我们已经开发了CLAMP（临床语言注释，建模和处理），一个临床 NLP工具，通过多个国际NLP挑战和大型 用户社区（来自700多个组织的用户下载了1 500多次）。CLAMP的商业化 Melax Technologies Inc.已经成功（即，现在有十几个持牌客户）;但它也揭示了其 作为云时代的桌面应用程序的限制。因此，我们建议将CLAMP扩展到一个新的云- 基于服务的平台（称为CLAMP-CS），它将通过以下方式解决已确定的挑战：1） 通过利用最先进的算法提高临床NLP性能并降低注释成本 例如深度学习、主动学习和迁移学习，并使经验不足的人也能使用这些方法， 2）遵循新的面向服务的架构，使CLAMP-CS通过SaaS和PaaS可用， 用于基于云的开发和部署;以及3）改进CLAMP-CS与下游的互操作性 应用程序遵循两种广泛使用的标准表示：HL 7 FHIR（快速医疗互操作性 资源）和OMOP CMD（公共数据模型），以支持临床操作和研究中的用例 分别有了这些先进的功能，我们相信CLAMP-CS将成为世界领先的临床NLP系统。 它将加速NLP技术在各种医疗保健应用中的采用， 临床/转化研究。

Project Summary Wide adoption of electronic health records (EHRs) has led to huge clinical databases, which enable the rapid growth of healthcare analytics market. One particular challenge for analyzing EHRs data is that much detailed patient information is embedded in clinical documents and not directly available for downstream analysis. Therefore, clinical natural language processing (NLP) technologies, which can unlock information embedded in clinical narratives, have received great attention, with an estimated global market of $2.65 billion by 2021 . In our previous work, we have developed CLAMP (Clinical Language Annotation, Modeling, and Processing), a clinical NLP tool with demonstrated superior performance through multiple international NLP challenges and a large user community (over 1,500 downloads by users from over 700 organizations). Commercialization of CLAMP by Melax Technologies Inc. has been successful (i.e., with a dozen licensed customers now); but it also reveals its limitations as a desktop application in the Cloud era. Therefore, we propose to extend CLAMP to a new Cloud- based, Service-oriented platform (called CLAMP-CS), which will address the identified challenges by: 1) improving clinical NLP performance and reducing annotation cost by leveraging the state-of-the-art algorithms such as deep learning, active learning and transfer learning and making them accessible to less experienced users; 2) following new service-oriented architectures to make CLAMP-CS available via SaaS and PaaS, ready for Cloud-based development and deployment; and 3) improving CLAMP-CS interoperability with downstream applications following two widely used standard representations: HL7 FHIR (Fast Healthcare Interoperability Resources) and OMOP CMD (Common Data Model), to support the use cases in clinical operations and research respectively. With these advanced features, we believe CLAMP-CS will be a leading clinical NLP system in the market and it will accelerate the adoption of NLP technology for diverse healthcare applications and clinical/translational research.