How good an algorithm do you need to take over your job?
This is a new question for many workers amid the rise of ChatGPT and other AI programs that can talk, write stories, and even generate songs and images within seconds.
But for doctors who examine scans to detect cancer and other diseases, more algorithms promise to improve accuracy, speed up work, and in some cases take over entire parts of the job. The arrival of AI has been looming for nearly a decade. Predictions range from doomsday scenarios where AI completely replaces radiologists to a brighter future where AI frees us up to focus on the most rewarding aspects of our jobs.
AI technology aims to detect breast cancer by mimicking radiologists’ eye movements: ‘an important friend’
This tension reflects how AI is being deployed across healthcare. Beyond the technology itself, much will depend on doctors’ willingness to place their trust and the health of their patients in the hands of increasingly sophisticated algorithms that few people understand.
There is disagreement within the field as to the extent to which radiologists should adopt this technology.
“Some of the AI technologies are so good that, frankly, I think they should be deployed right now,” said Dr. Ronald Summers, a radiologist and AI researcher at the National Institutes of Health. Told. “Why leave that information on the table?”
Dr. Laurie Margolies demonstrates the Koios DS smart ultrasound software on Wednesday, May 8, 2024, at Mount Sinai Hospital in New York. Breast imaging AI is used to obtain a second opinion for mammography ultrasound examinations. “I’m going to tell my patients, ‘I saw it, and the computer saw it, and we both agree,'” Margolies said. “I think patients feel even more confident when they hear that we are all the same.” (AP Photo/Mary Altafer)
Summers’ lab has developed computer-assisted imaging programs that detect colon cancer, osteoporosis, diabetes, and other conditions. None of these have been widely adopted, and he attributes this to, among other things, “medical culture.”
Radiologists have been using computers to enhance images and flag suspicious areas since the 1990s. But modern AI programs have gone even further, being able to interpret scans, provide diagnostics, and even generate reports on the results. The algorithms are trained using millions of his X-rays and other images, often collected from hospitals and clinics.
Across medicine, the FDA has approved more than 700 AI algorithms to assist physicians. According to recent estimates, more than 75% of them are engaged in radiology, but only 2% use such technology in radiology clinics.
Despite promises from industry, radiologists believe there are many reasons to be skeptical of AI programs. These include limited testing in real-world settings, a lack of transparency about how the AI works, and questions about the demographics of the patients used for training.
Artificial intelligence may not necessarily help reduce physician burnout, study suggests
“If we don’t know what cases the AI was tested on, or whether those cases are similar to the types of patients we see, we can’t know if this will work for us. That question is on everyone’s mind,” said Dr. Curtis Langlotz, a radiologist who runs the AI Research Center at Stanford University.
To date, all programs approved by the FDA require human involvement.
In early 2020, the FDA held a two-day workshop to discuss algorithms that can operate without human oversight. Shortly afterward, radiology experts warned regulators in a letter that they “strongly believe it is premature for the FDA to consider approval or clearance of such a system.”
But in 2022, European regulators approved the first fully automated software to review and generate reports on healthy, normal-appearing chest X-rays. Oxipit, the company that developed the app, has submitted a U.S. application to the FDA.
The need for such technology is acute in Europe, where some hospitals are facing months-long scan backlogs due to a lack of radiologists.
In the United States, that kind of automated screening is probably years away. AI executives say it’s not because the technology isn’t ready, but because radiologists aren’t yet accustomed to relying on algorithms to perform even routine tasks.
“We’re trying to get the message across that they’re over-treating people and wasting a huge amount of time and resources,” said Koios, which sells an AI tool for thyroid ultrasounds. said Chad McLennan, CEO of Medical. It’s not cancerous. “We tell them, ‘Let the machine take a look at it, sign the report, and that’s it.'”
Radiologists tend to overestimate their accuracy, McLennan said. Her company’s research found that doctors who viewed the same breast scan disagreed about whether to perform a biopsy more than 30% of the time. If the same radiologist saw her the same images a month later, he had a 20% chance of disagreeing with his own initial assessment.
According to the National Cancer Institute, about 20% of breast cancers are missed by routine mammography screenings.
And there is also the potential for cost savings. The average annual salary for radiologists in the United States is more than $350,000, according to the Department of Labor.
Experts say that in the near future, AI will function like an airplane’s autopilot system, performing important navigation functions but always under the supervision of a human pilot.
Dr. Laurie Margolies of New York’s Mount Sinai Hospital System says this approach provides peace of mind for both radiologists and patients. The system uses Koios Breast Imaging AI to obtain a second opinion for mammography ultrasound exams.
“I’m going to tell my patients, ‘I saw it, and the computer saw it, and we both agree,'” Margolies said. “I think patients feel even more confident when they hear that we are all the same.”
The first large-scale, rigorous trial testing AI-assisted radiologists versus solo radiologists offers hints at potential improvements.
Initial results from a Swedish study of 80,000 women show that one radiologist working with AI outperformed two radiologists working without the technology. It has been shown that mammograms can detect 20% more cancers.
In Europe, mammograms are examined by two radiologists to increase accuracy. But Sweden, like other countries, faces a labor shortage, with only about 70 breast radiologists in a country of 10 million.
Research shows that replacing a second reviewer with AI reduced human workload by 44%.
Still, the study’s lead author says it is essential that a radiologist make the final diagnosis in all cases.
Dr. Christina Lang, from Lund University, said that if automated algorithms miss cancer, it would be “very negative for trust in caregivers.”
The question of who is liable in such cases is one of the difficult legal questions that remains unresolved.
One result is that radiologists will likely continue to double-check all AI decisions to avoid being held responsible for errors. That could eliminate many of the predicted benefits, such as reduced workload and burnout.
CLICK HERE TO GET THE FOX NEWS APP
Only highly accurate and reliable algorithms will allow radiologists to truly detach themselves from the process, says Dr. Saurabh Jha of the University of Pennsylvania.
Until such a system emerges, Jha likens AI-assisted radiology to someone looking over their shoulder and constantly pointing out everything on the road and offering to help you drive.
“That’s not helpful,” says Jha. “If you want to help me drive, you can take over and let me sit back and relax.”
