What makes AI a friend, foe or time thief in radiology?
As artificial intelligence continues to expand across medical imaging, questions remain about whether the technology ultimately helps or hinders radiologists. According to Patricia Balthazar, MD, MPH, an assistant professor at the Emory University School of Medicine and medical director for quality and safety in radiology, the answer depends less on the technology itself and more on how it is implemented and managed.
“I’m not going to pretend that I have all the answers,” she said in the above video interview with Radiology Business. “The AI is not inherently a friend or foe or a time thief. It is what we make of it and how much we get involved and how the implementation works.”
Continuous monitoring of AI is critical to success
Balthazar stressed the importance of ongoing oversight after AI tools are deployed in radiology. While algorithms must demonstrate performance during regulatory review, she emphasized that real-world use can reveal new challenges. AI performance also can drift when imaging system parameters like slice thickness or other settings are changed, or if the patient population being imaged is different than the study population used to train artificial intelligence.
“When they were FDA cleared, they showed some data. But are they actually performing the same way in the real world? That is the data that we need to keep an eye on,” she explained.
She warned against treating AI deployment as a one-time milestone. Instead, it should be a continuous lifecycle requiring monitoring, governance and quality assurance.
"I would say the hardest part of AI is once it's implemented. It's the after part. All of the hype is about creating the AI and getting it FDA cleared It's almost like we're going up this mountain and we reach the peak and we celebrate, but we forget we still have the way down," she said.
That downward slope includes monitoring and the performance measures for quality assurance of these AI models.
Workflow changes can disrupt AI performance
Even small operational changes can disrupt AI systems, highlighting the need for vigilance. Balthazar shared a recent example from her institution where a server change altered image routing, preventing some AI tools from analyzing studies. Such issues underscore that AI systems are tightly linked to imaging workflows, including scanner settings, data routing and user access. Any change can introduce “drift” that affects performance.
“The job does not end once the AI gets to the practice, it continues,” she said.
Measuring the value of AI
Balthazar noted that evaluating whether AI actually saves time or improves outcomes requires deliberate effort at the institutional level. Her team is currently conducting analyses on AI’s impact, with results pending publication.
She also pointed to emerging national efforts, including registries from the American College of Radiology, aimed at collecting real-world performance data. These registries could allow institutions to benchmark AI tools and make more informed adoption decisions.
“If we have a registry that feeds the data automatically and continuously, then it will be the ideal case scenario,” she said.
Radiologists need to be in the driver’s seat with AI
For hospitals and imaging practices considering AI adoption, Balthazar stressed the importance of radiologist leadership. She cautioned against top-down decisions driven solely by administrative goals to appear technologically advanced. Instead, she advocated for structured governance models, such as multidisciplinary AI committees.
At Emory, an internal AI council evaluates new tools using standardized criteria and diverse expertise.
“The radiologists are the ones who have the expertise to judge and ask the right questions,” she said. “I would suggest having a committee…with different perspectives…to evaluate each individual software that comes along the pipeline.”
Ultimately, Balthazar framed AI not as a binary benefit or burden, but as a tool whose value depends on thoughtful integration and oversight. With proper governance, continuous monitoring, and collaboration across institutions, she suggested AI can deliver on its promise without becoming a hidden source of inefficiency.
“The AI is what we make of it,” she said.
Dave Fornell has covered healthcare for more than 17 years, with a focus in cardiology and radiology. Fornell is a 5-time winner of a Jesse H. Neal Award, the most prestigious editorial honors in the field of specialized journalism. The wins included best technical content, best use of social media and best COVID-19 coverage. Fornell was also a three-time Neal finalist for best range of work by a single author. He produces more than 100 editorial videos each year, most of them interviews with key opinion leaders in medicine. He also writes technical articles, covers key trends, conducts video hospital site visits, and is very involved with social media. E-mail: [email protected]