Chicago-based Northwestern Medicine has developed an AI tool that generates reports for radiologists. And the technology is already showing results — and saving the health system money.
For X-rays, the solution boosted radiologists’ productivity by an average of 15%, and up to 40% in some cases, without sacrificing accuracy, according to a June 5 study in JAMA Network Open.
“A lot of AI solutions are solving the wrong problem,” study coauthor Samir Abboud, chief of emergency radiology at Northwestern Medicine, told Becker’s. “The problem is not accuracy. The problem is not false positives and false negatives. The problem is that we don’t have enough radiologists to read the studies.”
The AI scans X-rays and CT scans to generate, almost instantaneously, reports that radiologists then review, edit and sign off on. The system also immediately flags a variety of life-threatening conditions like pneumothorax (aka collapsed lung) in real time; existing tools might detect only one condition at a time. The reports are 95% completed by AI and written in the individual radiologists’ style.
“With this generating the report for us in advance, our focus is on our patient,” Dr. Abboud said. “That has allowed us to operate at a much higher level as radiologists.”
Using the tool, Dr. Abboud said he now sees 40-50% more patients and feels less mentally drained at the end of the day. Engineers and clinicians worked side by side to develop the software, lessening the likelihood of hallucinations.
By developing its own solution, using an open-source platform from Meta, Northwestern also saved itself a lot of money. The health system trained the tool on millions of its own images.
“In terms of cost, it’s way less than any of the large language models you’re seeing out there,” said study senior author Mozziyar Etemadi, MD, PhD, clinical director of advanced technologies at Northwestern Medicine. “We’re talking 100 times less or even more.”
The tool won’t replace radiologists, said Drs. Etemadi and Abboud, but supplement and make their jobs easier, like an airplane autopilot. Adoption among Northwestern radiologists has been widespread, from recent residents to physicians nearing retirement.
“There’s no widely used physician extender in the radiology world,” Dr. Etemadi said. “There are no NPs or PAs, like an ER doc might have. AI is our NP or PA.”
Dr. Etemadi said the tool could eventually expand to other specialties, such as ophthalmology, dermatology and pathology. Northwestern is also working to commercialize the technology.
“It’s bigger than this project,” Dr. Etemadi said. “There is a way to do this better than the currently available ChatGPT-type things.”
