Radiologists Offer Advice on Successfully Advocating for Clinical AI

An estimated one third of all radiologists are using AI in the U.S., but the journey to get there requires some considerations, ranging from IT integration to workflow setup to department support. Below, radiologists who championed the integration of artificial intelligence, share their advice, covering:

  • Finding the use case and getting the hands-on experience
  • Identifying a narrative for administrators that supports adoption 
  • Building a coalition of innovation champions

Evan Kaminer, MD, former Director of Diagnostic Imaging, Montefiore Nyack

When you’re looking to integrate AI, the first step is to ask yourself: What’s your goal? What problem are you trying to solve? Because there are different types of AI and not all AI is the same. For example, there’s AI that helps you find lung nodules and there’s AI that helps you characterize the lung nodules. 

Radiologists should also ask about the whole workflow. How is the radiologist notified of the AI findings? Can it be integrated into the organization’s PACS? Does the radiologist have to use the tools provided by the AI vendor? Both?

Once you have a solution and get a framework of what you want to do, then it’s great to gain the experience using AI to solve that issue. It’s very hard to explain how AI helps unless you’re actually using it. At our institution, we also conducted some studies about AI accuracy and calculated its sensitivity, specificity and negative predictive value, which I think is the most important part of AI.

Dr. Kaminer’s Aidoc AI experience:

Glenn Garcia, MD, Director of Musculoskeletal Radiology, University of Texas Medical Branch

It’s in vogue to say, “Hey, let’s get AI in and go.” But you can end up with buyer’s remorse, if you don’t have a plan. Radiology departments have to ask: How’s this going to help us? Who is it going to benefit? Is it going to give us a sleepful night? Radiologists need to ask these questions, because the other departments, like the ER, won’t be asking us.

Start by pulling in the key players, such as a chest radiologist if it’s for pulmonary embolism; or if it’s a head bleed, your ER radiologist, your neuroradiologist. Get them on board and have a game plan as to what you want to solve. Make sure you’ve got the IT support, too, once you want to integrate.

Part of the game plan also means having a way to mark metrics to measure success and a story to tell that you can stand firm on, because once a department gets the initial funding, the team needs to show what the impact is to an administrator, especially for renewal contracts. It doesn’t necessarily need to be only a matter of lives saved, because that’s often difficult to show. But another narrative that is powerful is something to the effect of physician engagement—a “look who is using this tool” story. At our academic institution, it was about residents and that they relied on AI for as-needed elective imaging.

Dr. Garcia’s Aidoc AI experience:

Ryan Lee, MD, Chair of the Department of Radiology, Einstein Health Network

When a practice or network is interested in bringing AI algorithms into their workflow, it’s first important to test it outside of a production environment that has live patients. This way, you can see, in this test environment, how it’s going to work, what—if any—workflow limitations there are and understand exactly how this works in a live environment and try to work out all the issues before going live, including radiologist comfort.

Getting a coalition together of AI champions in the department is also a key to success, making sure the radiologists are onboard with it. This is essential for early adopters in the organization to be able to test the product out, because if you want your deployment of whatever the new technology is—whether it’s AI or some other technology—it’s important to have these advocates that can help drive the enthusiasm for adoption.

Dr. Lee’s Aidoc AI experience:

Explore the Latest AI Insights, Trends and Research