Many employees are asking new questions with ChatGPT and other AI systems that can create stories, have discussions, even generate melodies and graphics in a matter of seconds.
Though for the past ten years new algorithms promise to boost accuracy, speed jobs, and possibly take over whole work responsibilities, artificial intelligence has been a worry for doctors who review scans to find cancer and other diseases. Dark forecasts about the day when artificial intelligence replaces radiologists and optimistic futures when it lets radiologists focus on the most gratifying aspects of their work abound.
AI TECH TRIGgers Eye Movement of Radiologists to Detect Breast Cancer: “A Critical Friend”
That debate reflects the way the medical sector is using artificial intelligence. Beyond the technology in itself, much depends on doctors’ willingness to trust their patients’ health and their own confidence in increasingly more complicated algorithms few people know about.
Even within the field, opinions on the extent radiologists should be employing the technology vary.
Radiologist and AI researcher Dr. Ronald Summers of the National Institutes of Health said, “Some of the AI techniques are so effective, honestly, I think we should be doing them now. “Why are we only allowing that knowledge to sit there?”
Summers’ group developed computer-aided imaging technologies that can detect among other ailments diabetes, osteoporosis, and colon cancer. Among other factors, he says, the “culture of medicine” is the reason none of those have been generally embraced.
Radiologists have improved images and pointed up suspicious regions using computers since the 1990s. The most recent artificial intelligence systems, however, can examine the scans, generate a diagnosis, and even create reports on their results. Many times, the algorithms are taught using millions of X-rays and other images acquired from clinics and hospitals.
The FDA has approved over 700 artificial intelligence algorithms for usage in medicine to assist clinicians. Although over 75% of them work in radiology, just 2% of radiology practices, based on a recent estimate, use such technology.
Radiologists have various reasons to be wary of artificial intelligence programs despite business promises: insufficient real-world testing, opaque operations, and uncertainty regarding the patient demographics used to train them.
Studies show that artificial intelligence does not always help to reduce doctor burnout.
“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,” radiologist Dr. Curtis Langlotz of Stanford University said.
Currently authorized FDA initiatives all depend on human participation.
Early in 2020, the FDA called a two-day conference to discuss algorithms that might operate free from human oversight. Not long after, radiologists wrote to authorities stating they “strongly believe it is premature for the FDA to consider approval or clearance of such systems”.
But in 2022 European authorities approved the first completely automated software looking at and producing findings for normal and healthy chest X-rays. Software developer Oxipit is forwarding the FDA her U.S. application.
Europe really needs this type of equipment since some hospitals have months-long scan backlogs resulting from a shortage of radiologists.
Right now, that kind of automated screening is most likely years away in the United States. Not because the technology isn’t ready, assert artificial intelligence executives, radiologists aren’t yet comfortable assigning even basic tasks to algorithms.
“We try to tell them they’re overtreating people and they’re wasting a ton of time and resources,” said Koios Medical’s CEO, who provides an artificial intelligence tool for thyroid ultrasounds, most of which are not cancerous. “We tell them, let the machine view it, you sign the report and be done with it.”
McClennan says radiologists sometimes overstate their own accuracy. Research by his company found doctors who viewed the same breast photos disagreed with one another more than 30% of the time on whether or not to do a biopsy. Twenty percent of the time the same radiologists disagreed with their own original findings when they saw the same images a month later.
The National Cancer Institute thinks that 20% of breast tumors are missed in regular mammograms.
One might also be able to save money. American radiologists, according to the Department of Labor, often earn more than $350,000 annually.
Like aircraft autopilot systems, experts believe artificial intelligence will operate in the near future doing vital navigational duties under constant human pilot supervision.
According to Mount Sinai hospital system in New York’s Dr. Laurie Margolies, the approach not only comforts radiologists but also patients. The system gets a second opinion on mammography ultrasounds using Koios breast imaging artificial intelligence.
“I will tell patients, ‘I looked at it, and the computer looked at it, and we both agree,'” Margolies said. “I think that hearing me say we both agree gives the patient even more confidence.”
The first large, meticulous studies comparing radiologists utilizing artificial intelligence against those working alone provide hints of the many opportunities.
Preliminary results from a Swedish study involving 80,000 women show one AI-using radiologist found 20% more cancers in mammograms than two radiologists without the technology.
Two radiologists study mammograms throughout Europe in order to improve accuracy. Like other countries, Sweden suffers a personnel shortage; out of a population of 10 million, there are only roughly 70 breast radiologists.
The study indicated that substituting artificial intelligence for a second reviewer reduced the human load by 44%.
Lead author of the study does caution, though, that radiologists have to render the final diagnosis in every case.
“That’s going to be quite negative for trust in the caregiver,” Lund University researcher Dr. Kristina Lang said if an automated algorithm misses a cancer.
Among the challenging legal concerns yet unsolved is who would be responsible in such situations.
Radiologists will thus most likely keep double validating every AI discovery to avoid being held accountable for a mistake. Many of the expected advantages—less effort and burnout—probably will be lost as a result.
Professor Saurabh Jha of the University of Pennsylvania claims that radiologists could only fully disengage from the process using a very accurate and reliable algorithm.
Jha compares AI-assisted radiography to a person who offers to help you drive by checking over your shoulder and pointing out anything on the road until such technologies are built.
That is not helpful, Jha says. “If you want to help me, you take over the driving so that I may relax.”
