A new study from Washington University School of Medicine in St. Louis describes a novel method of analyzing mammograms that significantly improves the accuracy of predicting the risk of developing breast cancer over the next 5 years. Using up to 3 years of previous mammograms, the new method identified individuals at high risk of developing breast cancer 2.3 times more accurately than the standard method, which relies on questionnaires that assess only clinical risk factors, such as age, race and family history of breast cancer. The researchers showed that previous mammograms contain a wealth of information about early signs of breast cancer, that even a well-trained human eye cannot detect. This information includes subtle changes over time in breast density, which is a measure of the relative amounts of fibrous versus fatty tissue in the breasts.
A systematic review of artificial intelligence (AI)-guided mammography-based breast cancer prediction identified only three studies reporting a 5-year prediction horizon with AUCs between 0.68 and 0.70. One study reported higher performance for a 3-year or less prediction. Recent analyses of repeat digital images show that the temporal trajectory of density is significantly associated with long-term risk. However, the utility of longitudinal mammograms to dynamically predict women’s risk has not been characterized. For the new study, the team created an AI-based algorithm that can discern subtle differences in mammograms and help identify women at highest risk of developing new breast cancer in a specific time frame. In addition to breast density, their machine learning tool considers changes in other patterns in the images, including texture, calcification and asymmetry within the breasts.
Currently, risk-reduction options are limited and may include medications such as tamoxifen that reduce risk but can have unwanted side effects. In most cases, high-risk women are offered more frequent screening or the option to add another imaging method, such as an MRI, to try to identify cancer as early as possible. The researchers trained their machine-learning algorithm on mammograms from more than 10,000 women who received breast cancer screenings through Siteman Cancer Center from 2008 to 2012. They followed these individuals through 2020, and 478 were diagnosed with breast cancer during that time. The researchers then applied their method to predict breast cancer risk in a separate group of patients (more than 18,000 women who received mammograms from 2013 to 2020). Subsequently, 332 women were diagnosed with breast cancer during the follow-up period.
According to the new prediction model, women in the high-risk group were 21 times more likely to be diagnosed with breast cancer in the next 5 years than those in the lowest-risk group. In the high-risk group, 53 out of 1,000 women screened developed breast cancer in the next 5 years. In contrast, in the low-risk group, the figure was 2.6 out of 1000 women screened. With older questionnaire-based methods, only 23 out of 1000 women were correctly classified into the high-risk group. Full-body mammography images convey substantial information beyond breast density and summarize additional texture characteristics that are related to breast cancer risk independent of density. Cancer development is a slow, voluntary process that may be reflected in movement along pathways of morphological change in breast tissue, genetic changes and the immune environment.
The entire mammographic image can reasonably summarize the changes in the trajectory. Dynamic risk prediction methods have not been widely used. Instead, updating risk factors at each visit often supports repeated calculations of static risk based on the status at the current visit, without including previous history or trajectory of risk factors. The researchers demonstrate that the trajectory of previous mammograms adds to the performance of the model beyond a static approach. Additionally, they omitted patients with breast cancer diagnosed within 6 months of screening to ensure that the population of women for risk estimation and risk-related management was appropriate. Given the global burden of breast cancer among women, it is critical to use existing clinical image data to more accurately assess risk, guide personalized prevention, and better tailor screening strategies.
- Edited by Dr. Gianfrancesco Cormaci, PhD, specialist in Clinical Biochemistry.
Scientific references
Jiang S et eal. JCO Clin Cancer Inform. 2024 Dec; 8:e2400200.
Schopf CM, Ramwala OA et al. J Am Coll Radiol. 2024; 21:319.
Pashayan N, Antoniou A et al. Nat Rev Clin Oncol. 2020; 17:687.
