Entering the Real World of Artificial Intelligence

Integrating artificial intelligence into radiology is not as simple as it appears.


If you’ve ever used the search program on your smartphone to find all your photos of a special person or your pet, you are already using artificial intelligence (AI) for image interpretation.

While it looks like magic, AI programs that differentiate photos of cats from dogs, and even your own pet from others, are usually based on conventional mathematical optimization.

Under certain conditions, AI algorithms excel at detecting patterns in complex data, including patterns that humans just can’t see. They have an uncanny ability to extract valuable information from images, some of which we have been unable to obtain until recently.

Yet, as (relatively) easy as it is to build prototype AI tools to differentiate dogs from cats or even to distinguish genetic subtypes of tumors based on imaging features, it is difficult to translate them into robust, accurate, commercially viable, widely available products.

Even at the experimental prototype level, vagaries of medical data that challenge AI’s ability to translate image and electronic health record data into appropriate recommendations and best patient care are not completely understood.

Still, given its potential to identify image findings, triage cases, and perform other routine tasks, AI is expected to transform radiology more than any previous technological advancement since Wilhelm Roentgen discovered X-rays.

Statements like this tend to attract radiologists’ attention, as well as the attention of patients, entrepreneurs, companies, investors, government officials, and anyone else remotely involved with medical imaging.

But how do radiologists transition from curious, potentially apprehensive, observers of this change to experts with the skills and knowledge to guide the development of and manage this technology?

It’s a question ACR is helping to answer with its recently formed Data Science Institute (DSI). The institute is leading the creation of a national quality, technical, and leadership framework. Its intention is to define appropriate uses of AI in radiology; set standards for AI interoperability, quality, functionality, and ethical use; and address regulatory, legal, and economic issues associated with AI.

At the end of the day, though, the primary goal of the ACR DSI, and the College at large, is to educate radiologists about this emerging technology and provide the resources necessary to help radiologists excel in this new paradigm.

AI has come to radiology not as a magic, all-powerful robot that instantly replaces our current practice, but rather like puzzle pieces that must be joined together with our current workflow for success. To ensure everything fits seamlessly, radiologists must be involved, and ACR DSI will help.

By Raym Geis, MD, FACR, assistant professor of radiology at the University of Colorado School of Medicine.