Though AI is ingrained in American society, radiologists are reluctant to entrust an algorithm with their patients’ health.

Would an algorithm really need to be that good to replace you?

With ChatGPT and other AI systems that can write tales, have conversations, and even produce melodies and visuals in a matter of seconds, many workers are asking new questions.

AI has been a concern for physicians who examine scans to identify cancer and other illnesses, though, for the past ten years because new algorithms promise to increase accuracy, expedite tasks, and even take over complete job responsibilities. There have been dire predictions about the day when AI completely replaces radiologists and bright futures when it allows them to concentrate on the most fulfilling parts of their jobs.


That conflict is a reflection of the way AI is being implemented in the medical industry. Beyond the technology in se, a lot relies on physicians’ readiness to entrust their patients’ health and their own confidence to ever more complex algorithms that few people are familiar with.

Though views on how much radiologists should be using the technology vary even within the discipline.

Radiologist and AI researcher Dr. Ronald Summers of the National Institutes of Health remarked, “Some of the AI techniques are so effective, honestly, I think we should be doing them now. “Why are we just letting that information to sit there?”

Dr. Laurie Margolies walks through the Koios DS Smart Ultrasound program.
Dr. Laurie Margolies at Mount Sinai Hospital in New York on Wednesday, May 8, 2024, presents the Koios DS Smart Ultrasound program. Using artificial intelligence, mammography ultrasounds are second-opinioned. “I will tell patients, ‘I looked at it, and the computer looked at it, and we both agree,” Margolies said. “I believe that hearing me say that we both agree gives the patient an even higher degree of confidence.” (AP Image/Mary Altaffer)

Computer-aided imaging tools created by Summers’ group can identify diabetes, osteoporosis, and colon cancer among other diseases. He explains that among other things, the “culture of medicine,” is the reason none of those have been broadly accepted.

Since the 1990s, radiologists have enhanced images and indicated suspicious areas using computers. But the newest AI algorithms can analyze the scans, provide a diagnosis, and even write reports on their conclusions. Many times, the algorithms are taught using millions of X-rays and other pictures gathered from clinics and hospitals.

Over 700 AI algorithms to help doctors have been approved by the FDA for use in medicine. Despite the fact that over 75% of them work in radiology, just 2% of radiology practices, according to a recent estimate, use such equipment.

Despite corporate promises, radiologists have several reasons to be dubious of AI programs: little real-world testing, opaque workings, and uncertainties over the patient demographics used to train them.

Artificial intelligence not always helpful in lowering doctor burnout, research indicates

“There’s just a question in everyone’s mind as to whether these are going to work for us if we don’t know on what cases the AI was tested, or whether those cases are similar to the kinds of patients we see in our practice,” said Dr. Curtis Langlotz, a radiologist who oversees an AI research center at Stanford University.

All of the FDA-approved programs to date necessitate human involvement.

The FDA convened a two-day session early in 2020 to address algorithms that might function without human supervision. Not long afterward, radiologists wrote to authorities saying they “strongly believe it is premature for the FDA to consider approval or clearance” of such systems.

However, in 2022 European authorities approved the first fully automated program that examines and generates results for normal and healthy chest X-rays. Oxipit, the developer of the software, is sending the FDA its U.S. application.

Europe urgently needs this kind of technology since a lack of radiologists is causing months-long scan backlogs in some hospitals.

That kind of automated screening is probably years away in the United States right now. Radiologists aren’t yet comfortable delegating even simple chores to algorithms, not because the technology isn’t ready, claim AI executives.

CEO of Koios Medical, which offers an AI tool for thyroid ultrasounds, the great majority of which are not malignant, Chad McClennan stated, “We try to tell them they’re overtreating people and they’re wasting a ton of time and resources.” “We tell them, let the machine look at it, you sign the report and be done with it.”

McClennan claims radiologists often exaggerate their own accuracy. His organization conducted study that revealed doctors who saw the same breast images disagreed with one another more than 30% of the time about whether or not to perform a biopsy. When the same radiologists saw the identical pictures a month later, they even disagreed with their own original assessments 20% of the time.

In routine mammograms, the National Cancer Institute estimates that 20% of breast cancers are missed.

Savings might also be possible. The Department of Labor reports that American radiologists often make more than $350,000 a year.

Experts predict AI to operate in the near future much like airplane autopilot systems, carrying out crucial navigational tasks while always being overseen by a human pilot.

Dr. Laurie Margolies of the Mount Sinai hospital system in New York said the method provides comfort to patients as well as radiologists. Koios breast imaging AI is used by the system to obtain a second opinion on mammography ultrasounds.

“I will tell patients, ‘I looked at it, and the computer looked at it, and we both agree,” Margolies said. “I believe that hearing me state that we both agree gives the patient an even higher level of confidence.”

The first big, exacting studies pitting radiologists using AI against those working alone offer glimpses of the possible possibilities.

According to preliminary findings from a Swedish research including 80,000 women, one AI-using radiologist identified 20% more tumors in mammograms than two radiologists without the technology.

In Europe, to increase accuracy, two radiologists examine mammograms. But like other nations, Sweden has a workers shortage; out of a population of 10 million, there are only approximately 70 breast radiologists.

The study found that using AI in place of a second reviewer cut the human burden by 44%.

Lead author of the study however stresses that in every instance a radiologist must make the ultimate diagnosis.

“That’s going to be very negative for trust in the caregiver,” Lund University researcher Dr. Kristina Lang said if an automated program misses a cancer.

Among the difficult legal questions yet unresolved is who would be held accountable in such circumstances.

As a consequence, radiologists are probably going to keep verifying every AI finding twice in order to avoid being blamed for a mistake. Many of the anticipated benefits, such as less work and burnout, are probably going to be lost as a result.

According to University of Pennsylvania professor Saurabh Jha, radiologists could only completely detach from the process with an incredibly accurate and trustworthy algorithm.

Until such systems are developed, Jha compares AI-assisted radiography to a person who offers to assist you drive by watching over your shoulder and pointing out anything on the road.

That’s not useful, Jha remarks. “You take over the driving so that I can sit back and relax if you want to assist me.”


Related News