AI is revolutionizing disease detection, including breast cancer screening. New research from Lund University’s Masai trial shows that AI-supported screening can be significantly more effective than traditional methods.
Radiology uses imaging techniques like X-rays, CT scans, and MRI:s to diagnose diseases. Breast radiology relies on mammography, traditionally reviewed by two radiologists with Cad (Computer aided detection). In the Masai (Mammography screening with artificial intelligence trial) study, it is instead an AI software trained with deep learning that was used to screen 106 000 women.
The software assigns risk assessment scores from 1-10 to the images, so that only high-risk images require review by two radiologists, while remaining images only one. Recent results showed that AI-supported screening reduced the screen-reading workload by 44 percent and detected 29 percent more cancers, without resulting in more false positives. These are even more promising numbers than the first publication from 2023.
Research in breast cancer screening has a long and successful tradition in Sweden.
– We are often early in developing and implementing new methods and technologies, Kristina Lång, lead researcher in the study, explains.
This is thanks to researchers like Kristina Lång being able to work clinically while conducting studies, enabling efficient collaborations, she says. Compared to other cancer screenings, mammography also has a large patient group and a vast number of images that the AI algorithms can train on.
Development is also necessary, since there is a global shortage of radiologists, Kristina Lång explains. This shortage has already forced some countries to implement the technique, a progression that she thinks might have gone a bit too fast.
– Many advocate for using AI as the only tool – but I’m not too keen on that. I believe a human should be involved, and so do the women participating in the screening. But it’s very possible that it will happen, says Kristina Lång.
– I think it’s important to be cautious and have clear evidence when implementing something on a large scale.
Kristina Lång sees that AI is transforming the field of radiology, just as it reshapes society. However, she wants to point out that the study highlights medical expertise just as much as AI capability. From the first publication in 2023 to the most recent findings, the number of detected cancers increased from 20 percent more than normal to 29 percent more, although the AI algorithms did not change.
This increase is due to the radiologists’ ”learning curve”, where they learn to work with the machine more efficiently. Without a radiologist reviewing the screenings as well, there would also be many more false positives.
– It is hopeful that we have tools that can enhance human capabilities, but the radiologist remains at the center, Kristina Lång says.
The next phase of the Masai study will analyze interval cancers – those that develop between scheduled screening appointments – to further evaluate the AI system’s effectiveness. Kristina Lång and her team are hopeful about the results.
