A new image-only artificial intelligence model is reshaping how breast cancer risk can be predicted. During the annual RSNA meeting, researchers presented data showing that AI-based risk stratification provides significantly stronger and more precise prediction than breast density . This is a metric long used in routine screening but increasingly recognised as a limited indicator.
Traditionally, age, genetics and breast density guide decisions on who requires earlier or more frequent screening. Yet according to senior author Dr. Constance Lehman, Harvard Medical School, these markers are simply not sufficient. “More than two million women are diagnosed with breast cancer every year, often without warning. Only a small percentage is hereditary, and density alone is a very weak predictor of risk.” This is where image-based AI could transform preventive breast care.
A new generation of risk prediction
The AI model, Clairity Breast, the first FDA-authorised image-only risk predictor, was trained on 421,499 mammograms from 27 sites across Europe, South America and the United States. The dataset included scans from women who developed cancer within five years and women who remained cancer-free. This allowed the system to learn microscopic image patterns associated with elevated risk, many of which are undetectable to the human eye.
Using a deep convolutional neural network, the model was calibrated to predict individual five-year risk scores without relying on clinical or genetic history. It was then validated against more than 245,000 screening mammograms across six clinical sites. “AI can detect subtle structural variations in breast tissue that radiologists cannot see. This isn’t diagnosis, it’s prediction. And it opens the door to a completely new layer of screening intelligence”, said Lehman.
AI delivers clearer risk separation
Risk levels were categorised using standard National Comprehensive Cancer Network thresholds:
- Average risk: <1.7%
- Intermediate risk: 1.7–3.0%
- High risk: >3.0%
Women flagged as high-risk by AI developed cancer over four times more often than those identified as average-risk (5.9% vs. 1.3%). In contrast, breast density produced only marginal separation (3.2% vs. 2.7%).
In other words: AI didn’t just classify better, it stratified risk meaningfully. “Our large-scale analysis demonstrates that image-only AI provides far stronger predictive value than density alone. This supports more personalised and more equitable screening strategies”, according to First author Prof. Christiane Kuhl (RWTH Aachen University).
Earlier, tailored screening
Current American Cancer Society guidelines recommend that average-risk women may start screening at age 40. However, women under 40 represent the fastest-growing diagnosis group, and many have no known hereditary or clinical risk indicators. The ability to assign an AI-based risk score means clinicians could identify high-risk individuals long before conventional triggers appear.
“We already screen some high-risk women in their 30s based on genetics or family history. In the future, a baseline AI risk score at age 30 could help determine who should join that early screening pathway”, said Lehman.
The team suggests that future screening reports should incorporate both breast density and AI-derived risk, providing women a clearer understanding of personal risk rather than relying on a binary label. “We can do better than telling women their breast tissue is dense or not dense,” Lehman emphasises. “AI allows us to offer personalised, actionable information.”
Data-driven screening and prevention
This research represents an important shift in breast health: from reactive diagnosis to proactive prediction. With FDA authorisation and robust multi-regional validation, image-only AI may enable earlier detection among women with invisible risk and personalised screening schedules rather than age-based rules. Furthermore, more efficient resource allocation in national screening programmes improved outcomes through prevention instead of late-stage diagnosis
As breast cancer rates rise among younger women, AI-enhanced risk prediction could ultimately reshape screening policy and save lives. The next step is integration, ensuring that every mammogram is not just an image, but a measurable risk insight.
Interval breast cancer diagnosis
Earlier this year, researchers at UCLA Health Jonsson Comprehensive Cancer Center found that artificial intelligence can detect interval breast cancers earlier. These are tumors that appear between routine screenings and are often missed due to subtle imaging features.
In a retrospective review of nearly 185,000 mammograms (2010 to 2019), AI flagged cancers that were visible on the image but not recognized by radiologists. The team examined 148 interval cancer cases and categorized them using an adapted European system. Results suggest that integrating AI into screening could reduce interval breast cancer rates by up to 30%. Earlier detection may enable less aggressive treatment and improve patient outcomes, according to lead researcher Tiffany Yu.
