By Billy Candra 
A new study from Washington University School of Medicine in St. Louis has unveiled an innovative method for analyzing mammograms that promises to significantly enhance the precision of predicting an individual’s risk of developing breast cancer over the subsequent five years. This advanced approach, leveraging artificial intelligence to scrutinize up to three years of previous mammograms, has proven to identify individuals at high risk 2.3 times more accurately than the current standard, which relies primarily on questionnaires assessing clinical risk factors such as age, race, and family history of breast cancer. The findings, which represent a substantial leap forward in personalized cancer risk assessment, were published on December 5 in the esteemed journal JCO Clinical Cancer Informatics.
The Imperative for Enhanced Early Detection and Risk Stratification
Breast cancer remains the most frequently diagnosed cancer among women globally, posing a significant public health challenge. According to the World Health Organization, it accounts for approximately one in eight cancer diagnoses and is a leading cause of cancer-related deaths among women. Early detection is paramount to improving survival rates and treatment outcomes. Traditional breast cancer screening, primarily through mammography, has been instrumental in reducing mortality. However, the ability to accurately predict who will develop breast cancer, and when, has remained a complex challenge. Current risk assessment models, often based on questionnaires and broad demographic data, provide a general risk profile but frequently lack the granularity to identify subtle, early biological changes indicative of impending disease. This often leads to either over-screening for some women or, more critically, missed opportunities for intensified surveillance or preventative interventions for others who are truly at high risk.
The limitations of existing methods underscore the urgent need for more sophisticated tools. While factors like age, genetic predisposition, and reproductive history are known risk indicators, they do not capture the dynamic, evolving nature of breast tissue over time, which can harbor crucial predictive information. This gap in precise risk stratification has spurred extensive research into leveraging advanced technologies, particularly artificial intelligence and machine learning, to extract deeper insights from diagnostic imaging.
Unlocking Hidden Insights from Historical Mammograms
The study’s 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, emphasized the overarching goal: "We are seeking ways to improve early detection, since that increases the chances of successful treatment." Dr. Colditz also highlighted the broader implications for preventative medicine, stating, "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."
This innovative risk-prediction method is built upon foundational research previously led by Dr. Colditz and 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 earlier work demonstrated that prior mammograms contain a wealth of information regarding the early, subtle signs of breast cancer development—information that is often imperceptible even to the most highly trained human eye. This hidden data includes minute, temporal changes in breast density, a measure reflecting the relative proportions of fibrous versus fatty tissue within the breasts. Fluctuations in breast density, for instance, have long been recognized as a risk factor, but their precise quantification and predictive power in a dynamic context have been elusive until now.
For this latest study, the research team developed a sophisticated algorithm rooted in artificial intelligence, specifically machine learning. This algorithm is designed to discern these subtle, time-dependent differences across multiple mammogram images. Beyond breast density, the machine-learning tool meticulously analyzes changes in other critical patterns within the images, including variations in tissue texture, the emergence or evolution of calcifications, and alterations in breast asymmetry. As Dr. Jiang explained, "Our new method is able to detect subtle changes over time in repeated mammogram images that are not visible to the eye," yet these subtle shifts collectively hold rich, predictive information crucial for identifying high-risk individuals.
Rigorous Validation and Quantifiable Superiority
To develop and validate their groundbreaking algorithm, the researchers embarked on a comprehensive, multi-stage process. Initially, they trained their machine-learning algorithm using a vast dataset comprising mammograms from over 10,000 women who underwent breast cancer screenings through Siteman Cancer Center between 2008 and 2012. These individuals were meticulously followed through 2020, during which time 478 were subsequently diagnosed with breast cancer. This extensive training phase allowed the AI to learn and identify the intricate patterns associated with future cancer development.
Following the training, the algorithm’s predictive capabilities were rigorously tested on a separate, independent cohort of patients. This validation dataset included 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 validation group were diagnosed with breast cancer.
The results unequivocally demonstrated the new model’s superior accuracy. According to the AI-powered prediction model, women classified into the high-risk group were an astonishing 21 times more likely to be diagnosed with breast cancer over the subsequent five years compared to those in the lowest-risk group. To put this into perspective, the study found that 53 out of every 1,000 women screened in the high-risk category developed breast cancer within five years. In stark contrast, only 2.6 women per 1,000 screened in the low-risk group received a breast cancer diagnosis over the same timeframe. This represents a remarkable stratification of risk.
The study further highlighted the limitations of older methods. Under the traditional, questionnaire-based risk assessment, only 23 women per 1,000 screened were correctly identified and classified into the high-risk group. This discrepancy provides compelling evidence that the conventional approach, in this specific cohort, missed an estimated 30 breast cancer cases that the new AI-driven method successfully identified. This difference underscores the potential for the new algorithm to significantly reduce false negatives and ensure that more women who genuinely need intensified surveillance receive it.
Ensuring Equity and Broad Applicability
A critical aspect of the study’s design was its commitment to addressing health disparities and ensuring the algorithm’s robustness across diverse populations. The mammograms used for both training and validation were collected from a variety of settings, including academic medical centers and community clinics, thereby demonstrating the method’s consistent accuracy across diverse healthcare environments.
Crucially, the algorithm was developed 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 often led to models that perform less effectively in non-white populations, potentially exacerbating existing health inequities. Of the women screened through Siteman, a significant 27% were Black, while an even higher proportion, 42%, of those screened through Emory were Black. The researchers confirmed that the accuracy for predicting risk held up consistently across racial groups, a vital finding for equitable healthcare delivery.
The researchers are continuing their work, actively testing the algorithm in women from a broader spectrum of diverse racial and ethnic backgrounds, including those of Asian, Southeast Asian, and Native American descent. This ongoing effort aims to solidify the method’s universal applicability and ensure that it is equally accurate and beneficial for every individual, regardless of their ethnic heritage.
Transforming Clinical Practice and Prevention Strategies
The implications of this breakthrough are far-reaching, promising to reshape current breast cancer screening protocols and prevention strategies. As 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, articulated, "Today, we don’t have a way to know who is likely to develop breast cancer in the future based on their mammogram images." She added, "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."
Currently, risk-reduction options for breast cancer are somewhat limited. These can include medications such as tamoxifen, which can lower risk but are associated with potential unwanted side effects, making patient adherence a challenge. More commonly, women identified at high risk are offered increased screening frequency or the addition of supplementary imaging methods, such as MRI, to facilitate the earliest possible detection of cancer.
With the advent of this new AI-driven predictive model, clinicians will gain an unprecedented tool for personalizing patient care. For women identified as high-risk by the algorithm, intensified surveillance – perhaps through more frequent mammograms, supplemental MRIs, or even emerging imaging technologies – can be strategically implemented. This targeted approach could lead to earlier diagnoses for those who truly need it, potentially detecting cancers at more treatable stages. Conversely, for women identified as very low risk, the algorithm could potentially help reduce unnecessary screenings, alleviating patient anxiety and optimizing healthcare resource allocation.
Beyond screening, the enhanced risk prediction has profound implications for prevention research. By accurately identifying individuals who are highly likely to develop breast cancer, future studies can more effectively evaluate new preventative interventions, including lifestyle modifications, dietary changes, or novel pharmacologic agents, in precisely the population that stands to benefit most. This targeted approach could accelerate the discovery and validation of more effective and personalized prevention strategies, ultimately reducing the overall incidence of breast cancer.
The Future Landscape: Integration and Dissemination
The researchers are actively working with WashU’s Office of Technology Management to secure patents and licensing for this innovative method, with the ultimate goal of making it broadly accessible wherever screening mammograms are performed. Furthermore, Dr. Colditz and Dr. Jiang are exploring the formation of a start-up company dedicated to bringing this transformative technology to the wider medical community.
The journey from groundbreaking research to widespread clinical adoption involves several critical steps. Regulatory approval from bodies like the Food and Drug Administration (FDA) will be necessary to ensure the algorithm’s safety and efficacy in real-world settings. Integration into existing clinical workflows and electronic health records will require careful planning and collaboration with healthcare providers and technology developers. Addressing potential challenges such as data privacy, algorithmic bias, and the cost-effectiveness of implementation will also be paramount. However, the clear advantages in accuracy and personalized care strongly position this technology for significant impact.
This study represents a significant milestone in the broader context of artificial intelligence’s rapidly expanding role in healthcare. AI is increasingly being leveraged across various medical disciplines, from accelerating drug discovery and personalizing treatment regimens to improving diagnostic accuracy in fields like pathology and radiology. The Washington University team’s work exemplifies how AI can move beyond simple automation to uncover complex patterns in medical data that are beyond human cognitive capacity, thereby ushering in a new era of proactive, predictive, and personalized medicine.
The collaborative effort, funding from Washington University School of Medicine in St. Louis, and the pending patents related to predicting disease risk using radiomic images underscore the institutional commitment and innovative spirit driving this research. As the medical community moves towards more precise and individualized healthcare, this AI-powered mammogram analysis stands as a beacon of progress, promising a future where breast cancer risk is understood with unprecedented clarity, leading to more effective prevention, earlier detection, and ultimately, improved outcomes for millions of women worldwide.
Reference:
Jiang S, Bennett DL, Rosner BA, Tamimi RM, Colditz GA. Development and validation of a dynamic 5-year breast cancer risk model using repeated mammograms. JCO Clinical Cancer Informatics. Dec. 5, 2024.
