The American College of Radiology announced the formation of the Data Science Institute (DSI), an inter-disciplinary organization aiming to guide the implementation of artificial intelligence (AI) tools in radiology.
The DSI will define appropriate use cases, set standards for interoperability, validate algorithms and serve as a hub of discussion for important legal and ethical issues that surround AI in medicine.
AI has functioned as a sort of boogeyman in radiology the last few years—a threat to the profession just over the horizon that manifests in editorials and conference presentations. However, emphasizing the turf war aspect of AI undermines its usefulness, according to DSI Chief Medical Officer Bibb Allen, MD.
“Having someone tell you that a machine is doing your job in 10 years is something that can cause trepidation in a radiologist,” said Allen, who just wrapped up a one-year term serving as the President of the ACR. “Certainly, none of us believe that’s necessarily true, and it totally underestimates the value artificial intelligence can bring to medicine.”
The ACR will draw on experience gained shepherding radiology through the PACS transition, when they were heavily involved with the creation of the Digital Imaging and Communications in Medicine (DICOM) standard.
“That’s part of our role of looking out for patients—what standards do we use? I don’t think it benefits patients for everybody to go in a hundred different directions,” said Allen. “We had that with the Initial PACS implementation, before DICOM—which was revolutionary for our ability to adopt digital modalities.”
The DSI will propose and develop standards, leveraging contacts both in industry and in the U.S. Food and Drug Administration (FDA) to define a set of regulations that are truly interoperable. According to Allen, they also hope to build a shared repository for AI algorithms and software, similar to their Transmission of Imaging and Data (TRIAD) program for images.
TRIAD allows participating institutions to seamlessly exchange images and data for accreditation, clinical trials and registries, and Allen believes an AI-focused version is an important step in the maturation of the field.
“If you’re at an institution or you’re a developer, and you have your own data set that was created with your own equipment, and you test it on images that were developed on your system—all you really know for sure is how it works on your stuff,” he said. “Instead, we want to say this is truly an interoperable algorithm across all use cases.”
Although the DSI was just announced May 21, the college is already having meetings with the FDA and other industry stakeholders to help guide priorities, said Allen.
“If you look out there, we’ve got all these great algorithms coming down the pipe, but they haven’t been introduced into clinical practice yet,” said Allen. “Creating vehicles to get these things into practice is another focus for the DSI.”
The passage of the Mammography Quality and Safety Act in 1992 and the creation of the first appropriateness criteria in 1993 underscore the College’s advocacy expertise, demonstrating their ability to work with the FDA, Congress, and the industry as a whole.
“Our goal is not to just pay lip service. Our goal is to be in the trenches as a trusted partner for both industry and regulatory bodies,” said Allen.
AI technology is likely to change radiology, maybe drastically, but Allen never looks at it as a job-killer or as a threat.
“Think about all the information we generate from imaging exams, thousands of images per patient per exam of imaging data, “ said Allen. “I wonder, is there something embedded in there that even the most trained eye couldn’t see? These machine algorithms will tell us how to detect disease better—the machine versus radiologist picture isn’t really a warranted characterization. “