Artificial intelligence is able to accurately detect 20 percent more breast cancers from mammograms than traditional screening by radiologists, according to early results from a Swedish trial published overnight.

The study is the first randomized controlled trial to look at using AI in breast cancer screening and comes amid a dramatically shifting landscape for the technology and how it’s regulated.

The interim results, published in the Lancet Oncology late Tuesday, found that using AI-supported analysis of mammograms, alongside either one or two radiologists, was as good as using two radiologists without AI and led to 20 percent more cancers being detected. 

There was also a significant reduction in workload for radiologists, with the doctors having to spend 44 percent less time reading mammograms.

The trial, which is still ongoing, was conducted in Sweden and looked at over 80,000 women. Half the participants had two radiologists look at their mammograms without AI, while the other half had their mammograms analyzed by AI and a radiographer — except in cases where AI generated the highest risk score, in which case two radiologists assessed the screening.

The results come amid a significant uptick in interest around the potential opportunities and risks that AI poses for medicine and the world more broadly. AI is increasingly being deployed in medical settings, but there is also concern about how algorithms are trained and validated in these spaces, as well as the potential for bias and over-diagnosis. 

Meanwhile, the EU is planning stringent regulation around the use of AI and the European Medicines Agency is working on assessing the risks and benefits of the technology for drug development.

More to be done

Despite the positive findings, lead author Kristina Lång from Lund University in Sweden said the interim safety results “are not enough on their own to confirm that AI is ready to be implemented in mammography screening.” Lång and her colleagues are waiting for further results from the trial that will indicate whether using AI reduces the number of cancers detected between screenings and so if it is worth rolling out. 

“The greatest potential of AI right now is that it could allow radiologists to be less burdened by the excessive amount of reading,” said Lång. She said AI could eliminate the need for a second radiologist to read the mammogram, allowing radiologists to help more patients. 

Commenting on the results, Stephen Duffy, professor of cancer screening at Queen Mary University of London, said that reducing the burden on radiologists’ time was “an issue of considerable importance in many breast screening programmes.” But he said there could be concerns that using AI might over-detect harmless lesions.

As well as being the first randomized trial looking at using AI to help read mammograms, the study was one of the first to look at AI in the field of radiology. An editorial in European Radiology in January argued that randomized controlled trials were important for monitoring the safety of AI systems, which can produce “unpredictable and undetectable errors, not explainable by human logic.”