A framework to enhance radiology structured report by invoking NLP and DL: Models and Applications

Weill Medical Coll Of Cornell Univ, New York NY

Investigators

Linked publications & trials

Abstract

PROJECT SUMMARY/ABSTRACT In radiology practices, timely and accurate formulation of reports is closely linked to patient satisfaction, physician productivity, and reimbursement. While the American College of Radiology and the Radiological Soci- ety of North America have recommended implementation of structured reporting to facilitate clear and consistent communication between radiologists and referring clinicians, cumbersome nature of current structured reporting systems made them unpopular amongst their users. Recently, the emerging techniques of deep learning have been widely and successfully applied in many different natural language processing tasks (NLP). However, when adopted in a certain specific domain, such as radiology, these techniques should be combined with extensive domain knowledge to improve efficiency and accuracy. There is, therefore, a critical need to take advantage of clinical NLP and deep learning to fundamentally change the radiology reporting. The long-term goal in this appli- cation is to improve the form, content, and quality of radiology reports and to facilitate rapid generation of radiol- ogy reports with consistent organization and standardized texts. The overall objective is to use radiology-specific ontology, NLP and computer vision techniques, and deep learning to construct a radiology-specific knowledge graph, which will then be used to build a reporting system that can assist radiologists to quickly generate struc- tured and standardized text reports. The rationale for this project is that through integration of new clinical NLP technologies, radiology-specific knowledge graphs, and development of new reporting system, we can build au- tomatous systems with a higher-level understanding of the radiological world. The specific aims of this project are to: (1) recognize and normalize named entities in radiology reports; (2) construct a radiology-specific knowledge graph from free-text and images; and (3) build a reporting system that can dynamically adjust templates based on radiologists' prior entries. The research proposed in this application is innovative, in the applicant's opinion, because it combines deep learning, NLP techniques, and domain knowledge in a single framework to construct comprehensive and accurate knowledge graphs that will enhance the workflow of the current reporting systems. The proposed research is significant because a novel reporting system can expedite radiologists' workflow and acquire well-annotated datasets that facilitate machine learning and data science. To develop such a method, the candidate, Dr. Yifan Peng, requires additional training and mentoring in clinical NLP and radiology. During the K99 phase, Dr. Peng will conduct this research as a research fellow at the National Center for Biotechnology Information. He will be mentored by Dr. Zhiyong Lu, a leading text mining and deep learning researcher, and co- mentored by Dr. Ronald M. Summers, a leading radiologist and clinical informatics researcher. This application for the NIH Pathway to Independence Award (K99/R00) describes a career development plan that will allow Dr. Peng to achieve the career goals of becoming an independent investigator and leader in the study of clinical NLP.