For more than half a century, the mammogram has stood as the gold standard in the fight against breast cancer. However, its primary function has traditionally been reactive: identifying a tumor once it has already formed. New breakthrough research, funded by the Breast Cancer Research Foundation (BCRF), is shifting this paradigm from detection to prediction.
Recent findings published in the journal Radiology suggest that artificial intelligence (AI) can do more than just spot existing abnormalities; it can track subtle, invisible changes in breast tissue over several years. These AI-derived risk scores, which evolve over time, provide a "dynamic" roadmap of a woman’s future risk, often signaling the onset of disease up to six years before a clinical diagnosis is possible.
Main Facts: The Shift to Dynamic Risk Assessment
The core of this scientific advancement lies in the transition from static to dynamic risk assessment. Traditionally, a woman’s risk for breast cancer was calculated based on "fixed" or slowly changing variables: age, family history, genetic mutations (such as BRCA1 or BRCA2), and breast density. While useful, these factors are often insufficient. Remarkably, between 85% and 90% of women diagnosed with breast cancer have no significant family history and no known inherited genetic mutations.
The BCRF-funded study, led by Dr. Connie Lehman, a professor of radiology at Harvard Medical School and founder of Clairity Breast, introduces a more fluid approach. By using AI to analyze the intricate patterns within mammographic images—patterns often invisible to the human eye—researchers have identified "signals" that change as a woman ages or as her physiological state shifts.
Key Discoveries
- The "Invisible" Signal: AI models can detect microscopic textural changes in breast tissue that precede the formation of a mass.
- Trajectory Over Time: Unlike a single snapshot, the study emphasizes the trajectory of risk scores. A rising score over multiple years is a significant red flag, even if the individual mammograms are read as "normal" by a radiologist.
- Early Warning Window: Significant differences in risk trajectories between women who eventually developed cancer and those who did not were detectable as early as six years prior to diagnosis.
- Personalization: This technology allows for a "personalized" screening schedule, moving away from the one-size-fits-all annual or biennial mammogram for every woman.
Chronology: From Image Detection to Risk Prediction
The journey toward AI-integrated mammography has been decades in the making, but the last five years have seen an exponential leap in capability.
The Era of Computer-Aided Detection (CAD)
In the late 1990s and early 2000s, the first generation of AI, known as Computer-Aided Detection (CAD), was introduced. These systems acted as a "second set of eyes" for radiologists, highlighting suspicious areas on a mammogram. However, CAD was plagued by high false-positive rates and did not attempt to predict future risk; it only looked for what was already there.
The Development of Deep Learning (2015–2020)
With the advent of deep learning and neural networks, researchers began training AI on massive datasets of mammograms. Unlike CAD, which followed human-programmed rules, these deep learning models learned to identify complex features associated with cancer risk on their own. This led to the development of models that could provide a "five-year risk score" based on a single image.
The Move to Longitudinal Analysis (2021–Present)
The current study marks the latest chapter in this chronology. Dr. Lehman and her team recognized that a single score provides only a partial picture. By looking at a woman’s mammograms over a sequence of years, they could see how the AI’s perception of her tissue was changing. This longitudinal approach—the "dynamic" score—is what has proven so effective in identifying those on a path toward a cancer diagnosis.
Regulatory Milestones
The research has already transitioned from the lab to the clinic. The FDA recently authorized the "Clairity Breast" platform, an AI tool derived from this research that estimates a woman’s five-year risk. Furthermore, the National Comprehensive Cancer Network (NCCN) recently updated its guidelines to include AI-based mammographic risk assessment, recommending its use for women starting as early as age 35.
Supporting Data: Analyzing 160,000 Mammograms
The strength of the findings published in Radiology rests on the sheer volume of data analyzed. Dr. Lehman and her colleagues conducted a retrospective study of nearly 160,000 mammograms from a cohort of more than 54,000 women.
The Scoring Metric
The AI model used in the study generates a score that represents the likelihood of developing breast cancer within the next five years. These scores are typically represented on a numerical scale.
Comparing the Cohorts
The researchers divided the participants into two groups: those who eventually developed breast cancer and those who remained cancer-free. The data revealed a stark contrast in how their AI scores behaved over time:
- The Cancer Group: Among women who were eventually diagnosed, the median AI risk score was 2.1 at the five-to-six-year mark before diagnosis. However, as the date of diagnosis approached, the score rose steadily. By the final screening exam before the cancer was detected, the median score had jumped to 6.6.
- The Control Group: For women who remained cancer-free, the scores were remarkably stable. Throughout the same six-year period, their median scores fluctuated only slightly, ranging between 1.8 and 2.2.
The "Pre-Diagnosis" Surge
The data showed that the most pronounced increase in risk scores occurred during the two years immediately preceding a diagnosis. This suggests that the biological changes associated with early-stage oncogenesis create a "signature" in the breast tissue that AI can quantify long before a tumor is large enough to be seen or felt.
Official Responses and Expert Insights
The medical community has reacted with cautious optimism, viewing these findings as a bridge toward a more preventative model of healthcare.
Dr. Connie Lehman, the study’s lead investigator, emphasized the paradigm shift this represents. "We observed clinically relevant differences in risk trajectories between women who did and did not develop cancer," she stated. "Having a dynamic risk score opens up a whole new domain of more effective diagnosis and preventive therapies. Our findings demonstrate that image-based AI risk scores evolve over time and that changes in those scores may provide additional information about future breast cancer risk."
Dr. Lehman frequently draws a parallel between breast cancer risk and cardiovascular health. Just as a doctor monitors a patient’s blood pressure or cholesterol over several years to decide when to prescribe statins or lifestyle changes, radiologists can now monitor "breast health" via AI scores to decide when to escalate screening.
The Breast Cancer Research Foundation (BCRF), which funded the study, highlighted the importance of supporting high-risk, high-reward technology. BCRF leadership noted that traditional models miss too many women. By funding AI research, they aim to provide tools that work for the 85% of patients who do not have a "traditional" risk profile.
Clinical Adoption: The technology is already being implemented in major medical centers. Institutions like Beth Israel Deaconess Medical Center in Massachusetts and Invision Sally Jobe in Colorado have begun integrating Clairity Breast into their screening protocols, allowing patients to receive their AI risk scores alongside their standard mammography reports.
Implications: A New Era of Personalized Prevention
The implications of dynamic AI risk scores are profound, touching on everything from clinical guidelines to patient psychological well-being.
1. Moving Beyond "High Density"
For years, "dense breasts" have been the primary indicator for supplemental screening (like ultrasound or MRI). However, density is a subjective and often imprecise metric. AI risk scores offer a more nuanced assessment, identifying women with "non-dense" breasts who may still be at high risk due to other textural features detected by the algorithm.
2. Personalized Screening Intervals
Currently, the debate over whether women should be screened every year or every two years is ongoing. Dynamic AI scores could resolve this. A woman with a low, stable score might safely move to biennial screening, while a woman whose score has jumped from 2.0 to 4.0 in a single year would be flagged for immediate supplemental imaging or more frequent monitoring, regardless of her age or family history.
3. Early Intervention and Chemoprevention
If a woman is identified as "high risk" six years before a potential diagnosis, it opens a window for primary prevention. This could include lifestyle interventions (diet and exercise) or "chemoprevention" using medications like tamoxifen or aromatase inhibitors, which can significantly reduce the risk of developing hormone-receptor-positive breast cancers.
4. Reducing False Positives and Negatives
By understanding the "baseline" of a woman’s tissue through a dynamic score, AI can help radiologists distinguish between a benign change and a suspicious one. This has the potential to reduce the number of unnecessary biopsies (false positives) while ensuring that aggressive cancers are caught at their earliest, most treatable stage (reducing false negatives).
5. Patient Empowerment
"Knowledge is power" is a recurring theme in the study’s reception. Providing a woman with a quantifiable risk score allows her to be an active participant in her healthcare. However, experts caution that an elevated score is not a diagnosis. It is a tool for risk management, not a guarantee of future illness. Clear communication between clinicians and patients will be essential as these scores become a standard part of medical records.
Conclusion
The research conducted by Dr. Lehman and her team represents a milestone in the application of artificial intelligence to women’s health. By proving that breast cancer risk is not a static number but a dynamic, evolving trajectory, they have provided a new lens through which to view cancer prevention.
As AI-based risk assessment continues to expand into clinical settings across the globe, the mammogram is being redefined. It is no longer just a tool for finding what is wrong today, but a powerful instrument for predicting what might happen tomorrow—giving doctors and patients the precious gift of time to change the outcome.
About the Author
