By Billy Candra 
A new study from Washington University School of Medicine in St. Louis has unveiled an innovative method for analyzing mammograms that significantly enhances the accuracy of predicting breast cancer risk over a five-year period. This advanced approach, leveraging up to three years of an individual’s previous mammograms, has demonstrated a remarkable 2.3 times improvement in identifying high-risk individuals compared to the conventional standard, which relies solely on questionnaires assessing clinical risk factors such as age, race, and family history. The findings, which promise a paradigm shift in personalized breast cancer screening and prevention, were officially published on December 5, 2024, in the esteemed journal JCO Clinical Cancer Informatics.
The Imperative for Improved Risk Assessment
Breast cancer remains one of the most prevalent cancers globally, with millions of new diagnoses each year. Early detection is paramount to successful treatment, significantly improving survival rates and reducing the need for aggressive therapies. Current risk assessment models, while useful, have inherent limitations. These models often employ questionnaires that gather data on demographic factors (age, ethnicity), reproductive history (age at first period, pregnancies), personal medical history (prior biopsies, radiation exposure), and family history of breast cancer. While these factors provide a baseline, they offer a static snapshot and frequently fail to capture the nuanced, dynamic biological changes that precede cancer development. This can lead to both over-screening of low-risk individuals and, more critically, under-identification of women who are truly at an elevated risk but do not fit the traditional high-risk profile.
According to the American Cancer Society, approximately 1 in 8 women (13%) will develop invasive breast cancer in their lifetime. Despite advancements in screening technologies like mammography, a significant challenge persists in distinguishing between women who are likely to develop the disease in the near future and those who are not. This gap highlights the urgent need for more precise and personalized risk stratification tools, moving beyond population-level statistics to individual-level prediction. The Washington University study directly addresses this critical need by harnessing the vast, untapped information embedded within routine mammogram images.
Unlocking Hidden Insights: The AI-Powered Breakthrough
The new method is the culmination of extensive research led by senior author Graham A. Colditz, MD, DrPH, associate director of Siteman Cancer Center, based at Barnes-Jewish Hospital and WashU Medicine, and the Niess-Gain Professor of Surgery, alongside lead author Shu (Joy) Jiang, PhD, a statistician, data scientist, and associate professor of surgery in the Division of Public Health Sciences at WashU Medicine. Their prior work had already established that historical mammograms contain a wealth of subtle information regarding early signs of breast cancer progression, details that are imperceptible even to highly trained human radiologists. This includes minute, longitudinal changes in breast density—a critical measure indicating the relative proportions of fibrous versus fatty tissue within the breast, known to be an independent risk factor for cancer.
For this latest study, the research team developed a sophisticated artificial intelligence (AI) algorithm, specifically a machine-learning tool, designed to discern these subtle, evolving patterns across multiple mammogram images taken over time. Beyond breast density, the algorithm meticulously analyzes alterations in other crucial image characteristics, including breast texture, patterns of calcification, and any developing asymmetry within the breasts. These features, often overlooked or difficult to quantify consistently by human observers, are precisely what the AI is trained to detect and interpret.
"We are seeking ways to improve early detection, since that increases the chances of successful treatment," stated Dr. Colditz, emphasizing the clinical imperative. "This improved prediction of risk also may help research surrounding prevention, so that we can find better ways for women who fall into the high-risk category to lower their five-year risk of developing breast cancer." The potential for this tool extends beyond early detection, paving the way for targeted prevention strategies tailored to an individual’s specific risk profile.
Dr. Jiang elaborated on the algorithm’s unique capability: "Our new method is able to detect subtle changes over time in repeated mammogram images that are not visible to the eye, yet these changes hold rich information that can help identify high-risk individuals." This ability to see beyond the human visual threshold represents a significant leap forward in diagnostic imaging, transforming mammograms from a static screening tool into a dynamic predictive instrument.
Rigorous Training and Robust Validation Across Diverse Populations
To develop and validate their machine-learning algorithm, the researchers employed a rigorous, multi-stage process involving two distinct and large patient cohorts. The initial training phase utilized mammograms from over 10,000 women who underwent breast cancer screenings at Siteman Cancer Center between 2008 and 2012. These individuals were meticulously followed through 2020, during which period 478 women were subsequently diagnosed with breast cancer. This extensive dataset allowed the AI to learn and identify the intricate patterns indicative of future cancer development.
Following the training, the algorithm’s predictive power was rigorously tested on a separate, independent validation cohort. This group comprised more than 18,000 women who received mammograms through Emory University in the Atlanta area from 2013 to 2020. Over the follow-up period, which also concluded in 2020, 332 women in this cohort were diagnosed with breast cancer. The ability of the algorithm to perform accurately on an entirely new and distinct patient population underscores its generalizability and robustness.
The results from this validation phase were striking. According to the new AI-powered prediction model, women classified into the highest-risk group were an astounding 21 times more likely to be diagnosed with breast cancer over the subsequent five years than those in the lowest-risk group. To put this into perspective, among every 1,000 women screened, 53 in the high-risk category developed breast cancer within five years. In stark contrast, only 2.6 women per 1,000 in the low-risk group developed the disease during the same timeframe.
Critically, the study also highlighted the significant limitations of the older, questionnaire-based methods. These traditional approaches correctly classified only 23 women per 1,000 screened into the high-risk group. This means that the new AI method identified an additional 30 breast cancer cases per 1,000 screenings that the conventional method would have missed, demonstrating a substantial improvement in identifying truly at-risk individuals.
An equally important aspect of the study’s design was its commitment to diversity and equity. The algorithm was intentionally built with robust representation of Black women, a demographic group historically underrepresented in the development of breast cancer risk models. This oversight in past research has contributed to health disparities, as models trained predominantly on white populations may not perform as accurately in other racial and ethnic groups. Of the women screened through Siteman, 27% were Black, while the Emory cohort had an even higher representation of Black women, at 42%. The study proudly confirmed that the accuracy of the new method in predicting risk held up consistently across all racial groups, a crucial step towards reducing disparities in breast cancer outcomes.
In ongoing work, the researchers are extending their testing to women of even more diverse racial and ethnic backgrounds, including those of Asian, Southeast Asian, and Native American descent. This continuous effort aims to ensure that the method is equally accurate and beneficial for every individual, irrespective of their heritage.
Implications for Clinical Practice and Patient Care
The advent of this AI-driven risk prediction model carries profound implications for clinical practice, moving towards a more personalized and proactive approach to breast cancer management.
-
Personalized Screening Regimens: Currently, screening guidelines are largely age-based, recommending annual or biennial mammograms for women starting at a certain age. With a more accurate five-year risk prediction, screening could become highly individualized. Women identified as high-risk might benefit from more frequent mammograms, or the addition of supplementary imaging methods such as MRI, which is known to be more sensitive in certain cases but is also more expensive and resource-intensive. Conversely, those identified as very low-risk might safely space out their screenings, reducing unnecessary radiation exposure and anxiety. "Today, we don’t have a way to know who is likely to develop breast cancer in the future based on their mammogram images," noted co-author Debbie L. Bennett, MD, an associate professor of radiology and chief of breast imaging for the Mallinckrodt Institute of Radiology at WashU Medicine. "What’s so exciting about this research is that it indicates that it is possible to glean this information from current and prior mammograms using this algorithm. The prediction is never going to be perfect, but this study suggests the new algorithm is much better than our current methods."
-
Targeted Prevention Strategies: For individuals identified as high-risk, a proactive discussion about risk-reduction options becomes more pertinent. While current options are somewhat limited, they can include lifestyle modifications, genetic counseling, or even chemoprevention drugs like tamoxifen or raloxifene, which are known to lower breast cancer risk but may come with unwanted side effects. The ability to precisely identify who would benefit most from these interventions, and thus tolerate potential side effects for a significant reduction in risk, is a game-changer. This targeted approach could optimize the benefit-to-risk ratio for preventive treatments.
-
Reduced Patient Anxiety and Over-diagnosis: Improved accuracy in risk prediction can alleviate significant patient anxiety associated with uncertainty or false positives. By more precisely identifying who needs closer monitoring, the system can reduce unnecessary biopsies and follow-up procedures for low-risk individuals, optimizing healthcare resources and reducing the emotional toll on patients.
-
Addressing Health Disparities: The deliberate inclusion of diverse populations in the algorithm’s development and validation is a critical step toward ensuring equitable healthcare outcomes. Breast cancer mortality rates are disproportionately higher for Black women compared to white women, despite similar incidence rates. Part of this disparity is attributed to later diagnoses and less effective risk stratification. A truly accurate and universally applicable risk model can help bridge this gap, ensuring that all women receive appropriate and timely care.
The Broader Impact: Towards a Future of Proactive Health
This research stands at the forefront of the growing integration of artificial intelligence into medicine, particularly in diagnostic imaging. The field of "radiomics," which involves extracting high-throughput quantitative features from medical images, is rapidly evolving, and this study demonstrates its powerful potential to revolutionize disease prediction.
The broader implications extend to healthcare economics and public health. By optimizing screening protocols, healthcare systems could see a reduction in the costs associated with widespread, untargeted screening and the subsequent management of late-stage cancers. More importantly, from a public health perspective, a higher rate of early detection and targeted prevention could lead to a decrease in overall breast cancer mortality rates.
The researchers at Washington University are actively working with WashU’s Office of Technology Management to secure patents and licensing for this groundbreaking method. Their ultimate goal is to make this technology broadly available wherever screening mammograms are provided, transforming it from a research breakthrough into a widely accessible clinical tool. Drs. Colditz and Jiang are also exploring the founding of a start-up company dedicated to bringing this innovative technology to market, underscoring their commitment to translating their scientific discoveries into tangible benefits for patients worldwide.
This work, supported by the Washington University School of Medicine in St. Louis, represents a significant stride towards a future where breast cancer risk is not merely assessed but accurately predicted, allowing for truly personalized prevention and early intervention strategies that save lives. The pending patents for Jiang and Colditz related to predicting disease risk using radiomic images highlight the novel and proprietary nature of their innovative approach. As AI continues to mature, its role in healthcare will undoubtedly expand, with studies like this one laying the crucial groundwork for a new era of proactive, predictive medicine.