For radiology to truly benefit from AI’s potential, the practitioners must learn how to get the most information possible out of all available digital data.
That’s according to the authors of a new analysis published by the Journal of the American College of Radiology that described use cases critical for radiology to successfully evolve in the years ahead.
In 2013 and 2014, attendees at two joint retreats held by the RSNA Radiology Informatics Committee and American College of Radiology (ACR) Commission on Informatics worked to identify gaps in technology that may impose future limitations in our digital workflow.
“All agreed that to ‘bend the curve’ of radiology practice, the informatics community and vendors need to develop standard, easy-to-use tools to describe the data associated with the clinical and business practice of radiology.” wrote Marc Kohli, MD, department of radiology at biomedical imaging at the University of California, San Francisco, and colleagues, in the report.
According to Kohli and colleagues, the following four uses cases can help radiologists ‘bend the curve’ moving forward:
Registries can help keep such data in one centralized location, the authors noted, which is ideal for researchers looking to extract it when needed, assuming all parties agree on a set of required data elements.
They pointed out that the ACR hosts numerous registries that are helping advance patient care on a daily basis. The ACR Lung Cancer Screening Registry, for instance, allows providers who perform CT lung cancer screening for Medicare patients to submit their data manually online.
“Some of the elements for LCSR come from the radiology report (LUNG-RADS score, recommendations), and others come from the EHR (smoking history),” the authors wrote. “The structure and content of the radiology report, as well as smoking history, varies from site to site. Due to this variability in how the data are represented, the process of extracting, normalizing, and submitting data to a registry is left to individual sites to implement using a variety of strategies including vended solutions, manual chart review and web page submission, and custom middleware.”
Another relevant use case is research specifically focused on AI and how it can improve patient care, the report said. Institutions wanting to collaborate on AI research can often face a difficult decision: either focus on the study at hand and get super specific with the data you are looking for or keep things more simple, allowing any data you collect to be reused down the line for other studies.
AI Product Validation
“As more researchers and companies enter the diagnostic space with AI algorithms, there is an increasing need for independent validation,” the authors wrote. “This is especially important from a regulatory perspective. To effectively regulate AI algorithms, the FDA needs to be able to scientifically evaluate data regarding commercial marketing claims.”
Validating solutions requires a lot of reliable data, the team added, noting, “The data sets required for this validation have not yet been established, but would benefit from a shared data element definitions.”
Computer Assistance for Radiology Reporting
According to stakeholders, structured reporting remains one of the most popular topics in imaging informatics. Patients, referring physicians and radiologists themselves can benefit from an environment where one specialist’s report isn’t always completely different from another specialist’s report. Digital healthcare data and AI can work together to help get radiologists on the same page, which is also a key aspect of successfully implementing clinical decision support systems.
“Having a reusable framework to deliver and maintain decision support content would be valuable for the radiology community,” the authors wrote.