Medicine Artificial intelligence improves breast cancer screening according to study
According to a study, the medical use of artificial intelligence (AI) can improve breast cancer detection rates in a mammography screening program by almost 18 percent. The results exceeded the researchers' expectations.
15.01.2025, 06:52
SDA
The study by the University of Lübeck and the German start-up company Vara also found that the use of AI did not increase the rate of false-positive findings. The study, which was recently published in Nature Medicine, analyzed data from more than 460,000 women.
They all took part in the German breast cancer screening program at twelve screening sites between 2021 and 2023. Around half of the mammograms were analyzed using AI, while the other half were evaluated by traditional double assessment by radiologists.
More diseases detected
"Our first goal was to show that AI-based assessments are on par with human assessments," explained Alexander Katalinic, head of research and director of the Institute of Social Medicine and Epidemiology at the University of Lübeck. "However, the results exceeded our expectations: AI significantly improves breast cancer detection rates."
The study found that artificial intelligence identified 6.7 cases of breast cancer per 1000 women examined, compared to 5.7 cases per 1000 detected using conventional methods by 119 radiologists. This equates to one additional case of cancer per 1000 women screened.
Importantly, at the same time, the rate of women referred for further testing remained stable: 37.4 per 1000 for AI examinations compared to 38.3 per 1000 for conventional double assessments. This shows that the AI program did not lead to more false positives and therefore more stressful further testing with biopsies and the like.
Improved efficiency
Another important result of the study - the world's largest on this topic to date - is the potential of AI to improve the efficiency of breast cancer screening. Simulations suggest that breast cancer detection rates would still be 16.7 percent higher if all cases flagged as normal by AI were not reviewed by radiologists. In addition, the number of unnecessary recalls of patients for possible further examinations could be reduced by 15 percent, according to the University of Lübeck.
Any increase in the accuracy of the assessment of mammograms and any reduction in the workload of the radiologists carrying out the assessments would be beneficial. According to the University of Lübeck, 24 million individual images from mammography examinations have to be assessed every year in Germany alone.
DOI: 10.1038/s41591-024-03408-6
