AI-assisted mammograms cut risk of developing aggressive breast cancer

People screened for breast cancer by AI-supported radiologists are less likely to develop aggressive cancers before their next screening cycle than those screened by radiologists alone, raising hope that AI-assisted screening could save lives.

“This is the first randomized controlled trial of the use of AI in mammography screening,” says Kristina Lång of Lund University in Sweden.

The AI-based approach involves using the software – which was trained on more than 200,000 mammograms from 10 countries – to rank the likelihood of cancer being present in mammograms on a scale of 1 to 10, based on visual patterns in the scans. Exams receiving a score of 1 to 9 are then evaluated by an experienced radiologist, while exams receiving a score of 10 – indicating that cancer is most likely to be present – ​​are evaluated by two experienced radiologists.

A previous study found that this approach could detect 29 percent more cancers than standard screening, where each mammogram is evaluated by two radiologists, without increasing the rate of false detections – where a tumor is reported but follow-up tests reveal that it is not actually present or will not cause problems. “It was great,” says Fiona Gilbert of Cambridge University, who was not involved in the trial.

Now, Lång and his colleagues have discovered that the AI ​​approach also reduces the risk of developing so-called interval cancers. These are tumors that grow quickly between screenings – hence their name – and therefore tend to be particularly aggressive and more likely to spread elsewhere in the body.

Lång and colleagues made the discovery during an analysis of more than 100,000 women in Sweden, with an average age of 55. They randomly assigned about half of the women to undergo their standard series of breast cancer screenings, where each mammogram is evaluated by two radiologists. The remaining participants were first examined by the AI ​​model – developed by biotechnology company ScreenPoint Medical in Nijmegen, the Netherlands – and then the exams were evaluated by radiologists, most of whom had at least five years of experience analyzing mammograms.

Women who received AI-assisted screening were on average 12% less likely to develop interval cancer than women who received standard screening. “When we got the results, we were extremely delighted,” says Lång.

This result could be due to the fact that AI is better able to detect cancers at a very early stage. So while radiologists may overlook small tumors that could develop into interval cancer, AI can spot them.

Even so, the study was designed only to determine whether AI can work as well as standard screening, not to see if it can perform better, meaning more trials are needed to confirm that it is truly superior, says Lång.

Additionally, the team did not evaluate whether the AI-based approach worked better in certain ethnic groups. Further trials, including an ongoing trial in the United Kingdom, will help resolve this issue, Gilbert says.

Research should also be conducted to test whether less experienced radiologists see the same benefit when using AI, but Gilbert doesn’t expect a huge difference.

Building on these results, Lång expects the AI ​​approach to be rolled out in southwest Sweden, where the trial was carried out, within a few months. But it will likely take about five years for other countries to complete similar trials that justify deployment elsewhere, Gilbert says. “Countries need to see what the impact is on their own populations, where people are screened more or less often and are of different ethnic origins,” she says.

They also need to determine whether the AI ​​approach is cost-effective. By some estimates, AI support could be worth investing in if it reduces interval cancer rates by at least 5 percent. Radiologists will also need to be trained, although this probably won’t be too complicated because the software is quite simple to use, says Lång.

It’s important to note that even as AI improves, breast cancer screening should still involve radiologists, says Lång. “Women who participate in screening say they don’t want AI to be a standalone tool, they want a human to be involved, and I agree with them. I think it’s very important that it be a tool for radiologists,” she says.