Research Discovery:Can Artificial Intelligence Help Predict Breast Cancer Earlier?

What research shows about using mammograms and AI to estimate future breast cancer risk

What It Is

Researchers reviewed 41 studies that explored whether artificial intelligence, or AI, can use routine mammograms to predict a person’s future risk of developing breast cancer. The review included research published from 2012 through February 2025. Most studies were conducted in the United States and Europe.

Traditional breast cancer risk tools often use information such as age, family history, and breast density. Mammography-based AI takes a different approach by examining patterns in mammogram images, including details that may not be easily visible to the human eye. The goal is to identify people who may have a higher risk of developing breast cancer in the future.

How It Works

AI systems learn from large collections of mammograms linked to known health outcomes. The technology looks for image patterns associated with a later breast cancer diagnosis and uses those patterns to estimate a person’s risk over the next several years.

The studies assessed prediction periods ranging from about 1 to 10 years, with 5-year risk prediction being the most common. Most studies used standard 2D digital mammograms. Only three studies evaluated 3D mammography, also known as digital breast tomosynthesis.

Key Findings

Overall, the AI models had a moderate ability to separate people who later developed breast cancer from those who did not. Performance was similar whether the models predicted risk over 2 years, 3 to 4 years, or 5 years and longer.

Important limitations remain:

• All 41 studies looked back at previously collected data rather than testing AI during real-time patient care.
• White, non-Hispanic participants were overrepresented, making it unclear whether the models will work equally well across all communities.
• Only about one-third of the studies tested a model in a truly independent population.
• Only six studies assessed whether the level of risk predicted by AI closely matched the number of cancers that actually occurred.
• Some models appeared more accurate when studies included cancers already present at the time of the mammogram, rather than only cancers that developed later.

Community Impact

Mammography-based AI could eventually support more personalized breast cancer screening. People found to be at higher risk might benefit from additional imaging, shorter screening intervals, or discussions about prevention. People at lower risk might be able to avoid unnecessary tests.

However, these tools are not yet ready to guide screening decisions on their own. Researchers recommend testing them in real-world health care settings, with 3D mammograms, different imaging equipment, and populations that better reflect the diversity of the communities they are intended to serve.

The findings are promising, but careful evaluation is still needed to make sure AI risk predictions are accurate, useful, and fair for everyone.

Researcher: Celine M. Vachon, Ph.D. (REDI)