Well before ChatGPT reared its head and made AI a household topic around the world, the medical world was already keen on using AI-powered tools. For years, scientists have tested out whether bots could be used to help diagnose diseases in people, especially when it comes to the attention to detail required in assessing medical imagery. The results have been startling—even in some creepy ways.
So it seems AI will play a role in medicine moving forward, perhaps sooner rather than later. A new study published by The Lancet Oncology on Tuesday found that AI used in breast cancer screenings is remarkably accurate. AI-supported screenings successfully diagnosed 20 percent more instances of cancers when compared to the standard double reading of mammograms by two human radiologists. The use of AI did not increase false positive breast cancer diagnoses; and it in fact helped reduce the workload required in studying mammograms by a whopping 44 percent.
“These promising interim safety results should be used to inform new trials and program-based evaluations to address the pronounced radiologist shortage in many countries,” Kristina Lång, a breast cancer radiologist from Lund University in Sweden and the lead author of the new study, said in a press release. “But they are not enough on their own to confirm that AI is ready to be implemented in mammography screening. We still need to understand the implications on patients’ outcomes, especially whether combining radiologists’ expertise with AI can help detect interval cancers that are often missed by traditional screening, as well as the cost-effectiveness of the technology.”
ADVERTISEMENT
Mammography can help in diagnosing breast cancer early when it’s far more treatable—but it’s not perfect. Current estimates suggest that 20 to 30 percent of cancers that should have been spotted in earlier screenings are missed. False positives also occur frequently.
Even when mammography works, it requires a great deal of time and attention from radiologists who are specialized in reading mammograms. It can often take more than a decade for someone to be trained to properly read mammograms for signs of breast cancer.
Thus, some researchers believe AI could be used to alleviate that strain on resources. The new study is the first randomized trial to test out whether such a tool could help make breast cancer screenings safer and more efficient.
The paper is based on trials involving mammogram readings from 80,033 Swedish women, aged 40-80, between April 2021 and July 2022. Any women who had undergone a mammogram screening at four different health care sites in southwest Sweden during that time had their mammograms read by either a commercially available AI system designed to study mammograms (before also being read by one or two radiologists); or they were read exclusively by two trained radiologists (the control arm). Radiologists had final say on whether any women needed to be recalled for additional screenings due to suspicious findings.
AI-supported screening found one additional case of breast cancer for every 1,000 women screened than did the standard two-radiologist process. It also resulted in more recalls than helped lead to 41 more cancer diagnoses. Both methods resulted in the same 1.5 percent false positive rate.
The AI-supported method was also able to read more than 36,000 more readings than the standard approach; the researchers calculate it would have taken one radiologist assisted by AI 4 to 6 fewer months to read 40,000 examinations than the control arm in the same time.
Still, there are plenty of reasons to be cautious of the new findings. For one, all of the screenings happened in Sweden; it would be critical to learn how AI could operate on a much more diverse population of women (race and ethnicity information was not collected and used as part of any analysis of biases). The findings are also based on one single AI system, and cannot be generalized to other platforms.
And other studies have pointed out that AI could be a misleading crutch for some radiologists who aren’t exercising as much caution. In May, the journal Radiology published a study that found AI could impair decision-making by radiologists evaluating mammograms regardless of the physician’s level of expertise.
And AI systems are simply limited in the scope of how much they can evaluate based on a mammogram screening. In a comment published by the journal, Nerero Segnan, former director of Department of Screening at CPO Piemonte in Italy who wasn’t involved with the new study, highlighted an important question that lingers after the findings: “[I]s AI, when appropriately trained, able to capture relevant biological features—or, in other words, the natural history of the disease—such as the capacity of tumors to grow and disseminate?”
Nevertheless, there’s a clear need for more resources to help physicians make safer and faster diagnoses for breast cancer—and that paves the way for more study into AI could help. “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. “While our AI-supported screening system requires at least one radiologist in charge of detection, it could potentially do away with the need for double reading of the majority of mammograms easing the pressure on workloads and enabling radiologists to focus on more advanced diagnostics while shortening waiting times for patients.”
