RETURN TO ABSTRACT LISTING


Background
Natural language processing is a subset of artificial intelligence that process the information contained in raw text and applies this information to complete desired tasks. NLP has been used for many text-based applications such as online language translation, chat bots, and text prediction. Its utility in the field of radiology has also been well explored. Given that the primary method of communicating information in the field is the text-based radiology report, radiology is uniquely suited to benefit from advancements in NLP. In the past, applications such as information extraction from reports and cohort generation for research studies have been extensively studied. With the development of advanced NLP models such as transformers in recent years, advanced applications such as report simplification, quality assurance, and sentiment analysis are now being explored. Furthermore, efforts taken to improve patient access to health information (including radiology reports) through legislation, such as the 21st Century Cures Act, provide an opportunity for applications of NLP in radiology to improve patient outcomes by targeting patient literacy. In this exhibit, we give a brief history of NLP, discuss challenges in radiology that can be addressed by NLP and how they can be addressed by the latest advancements in NLP, and propose future steps to further integrate NLP in the practice of radiology.

Educational Goals / Teaching Points
The need for NLP in radiology. History and milestones. Current interest in NLP and state of the art applications of NLP in radiology. A review of the need for the following applications of NLP in radiology.

Key Anatomic/Physiologic Issues and Imaging Findings/Techniques
Challenges of current practices in radiology from the clinicians perspective (workflow, reporting errors) and patients perspective (lack of health literacy, availability of reports to patients and associated anxiety, poor follow up to recommendations). A brief review of the history of NLP and structure of current NLP models, focusing primarily on transformer based models. Exponential increase in interest in NLP as evidenced by 3-fold increase in radiological publications about NLP from 2015-2019. Focus on applications such as information extraction from reports, cohort generation for research applications, and labeling disease phenotypes. Opportunity for NLP to have clinical value in radiology through applications such as text simplification, text summarization, reporting accuracy/quality assurance, revenue loss, billing code generation, appropriate communication of disease acuity, and a special focus on structured reporting and NLP.

Conclusion
As NLP models continue to advance at a remarkable rate, novel applications in radiology will continue to be discovered. Furthermore the push for patient involvement in health care and changes in the field of radiology (widespread acceptance of report standardization) create ample opportunity and need for these advancements to be readily integrated into clinical practice.