By Billy Candra 
This groundbreaking research, published on December 5 in JCO Clinical Cancer Informatics, heralds a significant advancement in breast cancer early detection and personalized risk assessment. The study details an artificial intelligence (AI)-driven methodology that leverages up to three years of previous mammograms to identify individuals at high risk of developing breast cancer. This new approach demonstrated a remarkable 2.3 times greater accuracy in risk prediction compared to the current standard, which primarily relies on questionnaires assessing clinical risk factors such as age, race, and family history of breast cancer. The findings underscore a potential paradigm shift in how future breast cancer screening and prevention strategies could be tailored, moving beyond broad demographic and historical data to incorporate subtle, yet crucial, biological indicators discernible through advanced image analysis.
The Imperative for Improved Early Detection
Breast cancer remains one of the most common cancers among women worldwide, and while screening mammography has contributed significantly to early diagnosis and improved survival rates, its ability to predict future risk for individual patients has been limited. Current risk assessment models, often based on tools like the Gail model or Tyrer-Cuzick model, synthesize various personal and family medical histories, reproductive factors, and prior breast biopsy results to estimate risk. While valuable, these methods can sometimes overlook individual physiological nuances that might signal an elevated risk not captured by broad questionnaires. The inherent challenge lies in identifying women who, despite not fitting the typical "high-risk" profile, may still be on a trajectory toward developing the disease. Early detection is paramount, as it significantly increases the chances of successful treatment and improves long-term outcomes. However, a more precise understanding of an individual’s future risk could also pave the way for more targeted prevention strategies.
"We are seeking ways to improve early detection, since that increases the chances of successful treatment," stated 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. Dr. Colditz emphasized the dual benefit of this enhanced prediction, noting, "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."
Unlocking Hidden Information in Mammograms
The new risk-prediction method is the culmination of extensive past research 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 prior investigations revealed that historical mammograms contain a wealth of information about the early signs of breast cancer development that often elude even the most experienced human radiologists. This includes subtle, incremental changes over time in breast density, a measure of the relative amounts of fibrous versus fatty tissue in the breasts. While breast density is a known risk factor, its dynamic shifts over multiple years have been difficult to quantify and integrate into risk models effectively.
The team’s breakthrough lies in developing an algorithm based on artificial intelligence and machine learning capable of discerning these subtle, yet critical, differences across sequential mammogram images. This sophisticated tool moves beyond simply analyzing static images, instead evaluating patterns of change over time. In addition to changes in breast density, the machine-learning algorithm considers variations in other image characteristics, including breast tissue texture, the presence and evolution of calcifications, and any asymmetries within the breasts. These complex patterns, individually imperceptible to the human eye, collectively provide a powerful prognostic indicator when analyzed by AI.
"Our new method is able to detect subtle changes over time in repeated mammogram images that are not visible to the eye," explained Dr. Jiang, highlighting that these minute changes collectively hold "rich information that can help identify high-risk individuals." This ability to extract and interpret dynamic biological signals from imaging data represents a significant leap forward in medical diagnostics.
Rigorous Development and Validation
To develop and validate this innovative algorithm, the researchers undertook a comprehensive two-phase study. The machine-learning algorithm was initially trained on a robust dataset comprising mammograms from more than 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 diagnosed with breast cancer. This extensive longitudinal data allowed the AI to learn and identify the subtle radiographic patterns associated with subsequent cancer development.
Following the training phase, the algorithm’s predictive capabilities were rigorously tested on an independent, separate cohort of patients. This validation dataset included mammograms from over 18,000 women who received screenings 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 consistent performance across these distinct datasets underscores the algorithm’s robustness and generalizability.
Quantifiable Improvement in Risk Prediction
The results from the validation study were compelling, demonstrating a dramatic improvement in risk stratification. According to the new AI-driven prediction model, women classified into the high-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, the study found that 53 out of every 1,000 women screened in the high-risk category developed breast cancer over the next five years. In stark contrast, only 2.6 women per 1,000 screened in the low-risk group developed the disease within the same timeframe.
Comparing these figures to the older, questionnaire-based methods reveals the algorithm’s profound impact. The previous methods correctly classified only 23 women per 1,000 screened into the high-risk group. This indicates that, in this study’s context, the traditional approach potentially missed 30 breast cancer cases per 1,000 screened that the new AI method successfully identified as high-risk. This substantial reduction in missed high-risk individuals could translate into thousands of earlier diagnoses and potentially lives saved annually if widely implemented.
A critical aspect of the study’s design was ensuring the algorithm’s accuracy 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, demonstrating that the method’s accuracy holds up in diverse clinical environments. Importantly, the algorithm was built with robust representation of Black women, a demographic group that has historically been underrepresented in the development of breast cancer risk models. Of the women screened through Siteman, 27% were Black, while an even higher proportion, 42%, of those screened through Emory were Black. The study confirmed that the accuracy for predicting risk held up consistently across racial groups, addressing a crucial health equity concern in medical AI development. The researchers are continuing this important work, currently testing the algorithm in women of diverse racial and ethnic backgrounds, including those of Asian, Southeast Asian, and Native American descent, to ensure equitable accuracy for all populations.
Implications for Clinical Practice and Patient Management
The development of this highly accurate risk prediction tool carries significant implications for future breast cancer screening and prevention strategies. For women identified as high-risk by the AI, current risk-reduction options, while limited, could be more effectively deployed. These might include more frequent screening mammograms, the addition of supplementary imaging methods such as magnetic resonance imaging (MRI) for earlier cancer detection, or even chemoprevention drugs like tamoxifen, which can lower risk but may come with unwanted side effects. The ability to precisely identify who would benefit most from these interventions could help clinicians make more informed decisions, minimizing unnecessary treatments or anxieties for those at lower risk while intensifying surveillance for those who truly need it.
"Today, we don’t have a way to know who is likely to develop breast cancer in the future based on their mammogram images," said 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. Dr. Bennett expressed profound optimism about the research, stating, "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." This shift from reactive diagnosis to proactive risk management represents a fundamental change in clinical philosophy.
Beyond individual patient management, this technology could have broader impacts on public health. By enabling more personalized screening schedules, healthcare resources could be optimized, potentially reducing over-screening in low-risk populations while ensuring that high-risk individuals receive the most intensive and timely care. This could lead to a more efficient allocation of mammography slots, radiology expertise, and follow-up resources. Furthermore, the granular data provided by the AI could fuel further research into the underlying biological mechanisms of breast cancer development, potentially leading to the discovery of new prevention targets or therapeutic interventions.
Future Outlook and Commercialization
The researchers are actively working with WashU’s Office of Technology Management to pursue patents and licensing for this innovative method, with the ultimate goal of making it broadly available wherever screening mammograms are provided. This includes integrating the technology into existing mammography platforms and clinical workflows, ensuring it can reach a wide patient base. Dr. Colditz and Dr. Jiang are also exploring the establishment of a start-up company focused on commercializing and disseminating this vital technology. This dual approach of academic publication and commercial development underscores the team’s commitment to translating their research from the lab into tangible benefits for patients worldwide.
The work was supported by Washington University School of Medicine in St. Louis, and Drs. Jiang and Colditz have patents pending related to this work, specifically concerning the prediction of disease risk using radiomic images. This commitment to intellectual property protection is crucial for ensuring the responsible and widespread adoption of the technology, facilitating further research, and maintaining the quality and integrity of the diagnostic tool as it moves into broader clinical use. As AI continues to integrate into medical diagnostics, this study stands as a testament to its potential to revolutionize early disease detection, offering a more precise, proactive, and personalized approach to healthcare.
