Kevin L. Smith, MD, sees AI changing the practice of radiology. “It’s kind of like magic — that’s how one of our radiologists described it,” he said.

How Radiology is Becoming a Leader in Adopting AI

Few clinical areas have adopted AI tools faster than radiology, easing workloads and helping overcome a shortage of clinicians.

Bobby King Dec 13, 2024

IN A DIM LIT room in the basement of University Hospital, Kevin Smith, MD, sits in front of a bank of video displays, a computer mouse in one hand, a dictation wand in the other.

Before him is a CT scan of a patient’s chest. Actually, it is a composite created by 656 separate images rolled into one, producing a 3D sketch of the patient’s heart and lungs.

As Smith moves his cursor across the screen, he can look at the lungs from different perspectives. And, within the ghostly black-and-white image is a cloud of blue, calling his attention to a spot on one of the lungs. It’s one of the AI programs he’s using that has identified as a suspicious lung nodule.

Is it cancer? Something harmless? For now, the trained human eye of the radiologist concludes that it only needs further monitoring, and more images down the road to look for changes. But the exercise is a glimpse into the new world of medical imaging that’s being greatly enhanced by artificial intelligence.

No clinical area has adopted artificial intelligence as quickly as radiology and imaging sciences. Of the nearly 1,000 AI and machine learning devices the FDA has authorized for health care, roughly 75 percent are for purposes in medical imaging.

“It is changing how we practice radiology,” said Smith, an assistant professor of clinical radiology and imaging sciences at IU School of Medicine who leads the clinical AI program in radiology at IU Health. “It plays to our strengths and lets us do the things we’re good at and takes away the things we are not good at, such as finding tiny dots on a screen over hundreds of images.”

And there are other AI tool at Smith’s disposal. First, a program takes dictation as he describes what he sees on a spinal X-ray. But it doesn’t make an audio record, it inserts them into his radiology report in real time — the words literally appearing on the screen seconds after he utters them. When Smith finishes the narrative, he gives the AI a command that prompts the technology to write his conclusion for him — a summary of the most notable things he saw in the images. What would take several minutes of his time, gets done in a couple of seconds. “It’s kind of like magic — that’s how one of our radiologists described it,” Smith said.

It even catches things that don’t add up, such as if the radiologist reports seeing right side renal tumor but then later refers to it being on the left. “We don’t like to talk a lot about human errors in medicine, but they happen,” Smith said. “Those kinds of errors are completely eliminated with this technology.”

“Computers can look at a billion images without fatigue. But when it finds something, it may struggle to figure out what it has found. In contrast, radiologists are experts at determining whether something is just a dot or potentially cancer.”

Kevin L. Smith, MD

The impact of AI in radiology is magical considering the landscape in the profession.

The physician shortage is widespread, but it is becoming acute in radiology — a factor of both the graying workforce and a relatively stagnant number of radiology residents, all happening as the population ages and needs more imaging.

Then there is the steep workload. At any given time, depending on their location, a radiologist may have anywhere from 50 to 100 cases a day awaiting their attention. A single case may require reviewing hundreds of images or, in the case of an MRI study, as many as a thousand.

Before AI, the search for lung nodules that Smith reviewed quickly with the 3D composite would have required him to review 656 separate images. “That’s what makes AI valuable. How long does it take for a human to get through that?” Smith said. “If you make that easier, you get real incremental improvements. Imagine if you’re reading 60 of these a day, when every scan is marked for them, that gets to be a pretty significant payoff.”

The possibilities in medical imaging are broad.

At its breast centers, IU Health has conducted trials for AI algorithms that can read mammograms, a practice Smith says is common in Europe. The goal is to eventually have every mammogram reviewed by AI. The advantage is obvious: A second reader — either the AI or the human could go first — may pick up signs of small cancers or changes in tissues the other might miss.

IU radiologists are also using a generative AI application that relieves the physician of the responsibility of ensuring a patient gets proper follow-up care. The software reviews the reports and flags them so a non-physician member of the staff can arrange for the patient’s follow-ups to be scheduled.

In the end, AI promises to help radiologists work more efficiently, avoid “major misses” of significant findings, and give them more time for the important stuff.

As exciting as the technology is, Jason Allen, MD, PhD, who is chair of the Department of Radiology and Imaging Sciences, said there’s no replacing the decision making of humans — even in an AI-ready field such as imaging.

“Computers can look at a billion images without fatigue,” he said. “But when it finds something, it may struggle to figure out what it has found. In contrast, radiologists are experts at determining whether something is just a dot or potentially cancer.”

With that in mind, first- and second-year radiology residents at IU are still learning to read images the old-fashioned way — one at a time — and to draft their own conclusions. Then, they get to try the newer technology.

Donor support can help the Department of Radiology attract talented researchers and back bold research. Learn more about its priorities and make a gift today.

Glossary Terms

Artificial General Intelligence: AI models with a level of intelligence on par with a human. While it has not been achieved, cognitive scientists think it’s not far off.

Algorithm: A set of instructions spelling out steps that a computer – or person – executes to solve a specific problem or task, such as pattern recognition.

Tags: Artificial Intelligence, Radiology