AI is coming to medicine, but it’s got a lot to learn

When you’re a journalist, potential sources are everywhere, even at a medical appointment. When I had my annual mammogram in May, radiologist Pouneh Razavi said all looked well. Then she said: “And we used AI!”

We here at Science News had been talking about medical uses of AI, so I had to find out what was up. It turns out that Sibley Memorial Hospital in Washington, D.C., which is part of Johns Hopkins Medicine, had started using artificial intelligence to help read mammograms just two months before. Razavi, who is director of breast imaging in the National Capital Region, Johns Hopkins, told me that my scan and others would be used to train algorithms, with the goal of giving more information to physicians and making diagnoses more accurate. I had done my bit for science while wearing a cotton robe.

My experience will almost certainly become increasingly common. Artificial intelligence is already in use in medical care, and many of those efforts focus on medical imaging (SN: 12/17/22 & 12/31/22, p. 32). Razavi isn’t the only medical professional excited about the possibilities; when Science News senior writer Meghan Rosen attended a conference on AI and medicine in New York last month, she found that at least half of the presentations were about medical imaging.

“There was an attitude of excitement and hopefulness,” Rosen told me. “It really interested me that one field is leaping ahead, and I wanted to know why.”

Rosen learned that radiologists have to perform detailed analyses of large numbers of scans, day after day after day. Although radiologists’ daily error rates are fairly low — Rosen says it averages 3 to 5 percent — errors can creep in when the doctors are overworked. AI could help by learning to recognize signs of disease, alerting the physician and helping them prioritize the most serious cases.

After the conference, Rosen talked with Razavi, the radiologist who told me about my AI-assisted mammogram. I had told Rosen about my experience because I thought it was pretty darn cool, and because I thought the doctor might be willing to talk with Rosen about her work experimenting with a new technology in the clinic.

There’s more than enough hype about the potential for AI to change the world, and more than enough reason for skepticism. Rosen says the same physicians and researchers who are excited about the possibilities of AI to improve medical imaging are also cautious; these are early days. “It was hopeful to see that this was not being hyped,” she says. “Scientists were talking about how difficult it is to get it working well in the real world.”

AI will not become a tool that reveals every detail about a person’s health in a scan; that’s going to remain in the realm of science fiction for the foreseeable future. But I too am optimistic that the technology can make complex tasks like reading mammograms more accurate, and free up more of doctors’ precious time. It’s not a magic wand, but a tool that may help physicians better serve their patients.

Nancy Shute is editor in chief of Science News Media Group. Previously, she was an editor at NPR and US News & World Report, and a contributor to National Geographic and Scientific American. She is a past president of the National Association of Science Writers.