How AI Helped Radiologists Detect More Breast Cancers
In a large real-world breast cancer screening study, radiologists working with artificial intelligence detected more cancers without increasing the number of women recalled for additional testing. The results suggest that AI can strengthen mammography screening—not by replacing radiologists, but by helping them recognize findings that deserve a closer look.
What the study found
Researchers studied 463,094 women between the ages of 50 and 69 who participated in organized mammography screening at 12 sites in Germany. Of those screenings, 260,739 were reviewed with AI support. A total of 119 radiologists participated, choosing whether to use the AI-supported system.
Radiologists using AI support detected 6.7 breast cancers per 1,000 screenings, compared with 5.7 per 1,000 without it—a 17.6% higher detection rate. The recall rate, which measures how often women were asked to return for further examination, did not increase.
How the partnership worked
The AI system did not make the final diagnosis. It helped organize the radiologists’ attention. It could identify examinations it considered normal and flag potentially suspicious findings for closer review.
One feature acted as a safety net. If the AI detected a suspicious area in an examination that a radiologist had initially assessed as normal, it highlighted the area and invited reconsideration. Radiologists could also override the AI—and they did. In one example reported by the researchers, both readers rejected the AI’s “normal” classification and correctly pursued a cancer diagnosis.
The meaningful advance was not AI working alone. It was the combination of computational pattern recognition and experienced clinical judgment.
What this does – and does not – prove
The findings are encouraging, but they are not proof that every AI system will improve every breast cancer screening program. This was a prospective, real-world observational study—not a randomized clinical trial. The participating radiologists chose whether to use the AI-supported system, and the research took place within Germany’s organized screening program for women ages 50 to 69. The results may not apply in exactly the same way to other populations, healthcare systems, or AI tools.
The study was funded by Vara, the company that developed the AI system. Vara employees participated in the study’s design, data collection, interpretation, and reporting, and several authors disclosed financial or professional relationships with the company. These facts do not invalidate the results, but they matter when weighing the evidence. Independent research and longer-term patient outcomes will help determine how broadly the findings should be applied.
Why this matters
Mammography screening asks radiologists to examine an enormous number of images while remaining alert to subtle signs of cancer. Used responsibly, AI may help direct attention toward findings that warrant another look while allowing clinicians to retain authority over the decisions that follow.
This is the kind of progress Bright AI Horizons follows: technology extending what people can detect, while human expertise, accountability, and judgment remain essential.