AI-driven breast cancer screening ‘a long way off’ from replicating radiologists
Breast screening programs are best served using radiologists to read exams, as new evidence shows artificial intelligence isn’t ready for the bright lights just yet.
Researchers pored over 12 studies performed since 2010 involving more than 130,000 women screened across the U.S., Sweden, Germany, the Netherlands and Spain. Ninety-four percent of AI systems were less accurate than a single radiologist, and all fell short when put up against the consensus of at least two rads.
The full results were published Sept. 2 in the bmj.
“Current evidence on the use of AI systems in breast cancer screening is a long way from having the quality and quantity required for its implementation into clinical practice,” Karoline Freeman, a senior research fellow at the University of Warwick’s Division of Health Sciences in the U.K., and colleagues explained.
Three well-rounded studies with nearly 80,000 women compared AI tools against radiologists’ decisions. Of this group, 1,878 had cancer spotted during screening or between routine appointments within 12 months of a breast exam.
While most programs came up short, five small studies involving approximately 1,000 women revealed that each AI system beat out a single radiologist. These investigations, however, proved to be at risk of bias, and the results were not replicated in larger studies.
In a handful of trials, AI commissioned to pre-screen and triage mammograms for radiologist review accurately disregarded low-risk cases but also tossed out cancers spotted by imaging providers.
Given their findings, Freeman et al. believe artificial intelligence tools may have a long road ahead.
“The findings from our systematic review disagree with the publicity some studies have received and opinions published in various journals, which claim that AI systems outperform humans and might soon be used instead of experienced radiologists,” the authors cautioned. “Although a great number of studies report the development and internal validation of AI systems for breast screening, our study shows that this high volume of published studies does not reflect commercially available AI systems suitable for integration into screening programs,” they added later.
Matt joined Chicago’s TriMed team in 2018 covering all areas of health imaging after two years reporting on the hospital field. He holds a bachelor’s in English from UIC, and enjoys a good cup of coffee and an interesting documentary.