Longitudinal AI analysis identifies changing risk patterns
Researchers have found that AI-generated breast cancer risk scores derived from screening mammograms evolve differently over time in women who later develop breast cancer than in those who remain cancer-free. The findings, published in Radiology, suggest that longitudinal analysis of image-based AI risk scores could support more personalized breast cancer risk assessment.
The study evaluated serial screening mammograms from women examined between 2009 and 2019 at six imaging sites representing urban, community, and rural practices. Researchers analyzed 158,807 mammograms from 54,014 women using a validated, open-source image-only deep learning model that generated continuous five-year breast cancer risk scores. The model relied exclusively on mammographic images without incorporating demographic, clinical, or historical imaging data.
Among the study population, 817 women were diagnosed with invasive breast cancer or ductal carcinoma in situ (DCIS), while 53,197 women remained cancer-free during follow-up.
Risk scores increased years before diagnosis
The researchers found that AI-derived risk scores increased progressively during the six years preceding a breast cancer diagnosis. Median scores rose from 2.1 during the first five to six years before diagnosis to 6.6 at the index examination. In contrast, women who did not develop breast cancer maintained stable scores, with median values ranging from 1.8 to 2.2 throughout the study period.
"We observed clinically relevant differences in risk trajectories between women who did and did not develop cancer," said lead author Constance D. Lehman, M.D., Ph.D., professor of radiology at Harvard Medical School and CEO of Clairity, Inc. "The increase in scores among cancer patients was detectable as early as six years prior to diagnosis and became more pronounced over time."
The study also showed that the increase in risk scores accelerated during the two years immediately preceding diagnosis, whereas cancer-free participants demonstrated essentially unchanged trajectories throughout follow-up.
Potential role in personalized screening
According to the researchers, the findings support the potential use of image-based AI risk models as dynamic imaging biomarkers for breast cancer risk assessment.
"These findings demonstrate signals, invisible to the human eye, in the image alone can predict future risk," Dr. Lehman said. "This is exciting, because 85% of women diagnosed with breast cancer do not have a significant family history of breast cancer or known genetic mutations."
The researchers also reported that the observed trends remained consistent across patient subgroups defined by age and breast density.
"AI-derived risk scores can identify patients who are otherwise predisposed to the disease, and 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 said.
The study authors noted that AI image-based risk scores have been incorporated into the 2026 National Comprehensive Cancer Network guidelines. An FDA-approved AI-based five-year breast cancer risk-scoring model is also currently in clinical use at select healthcare institutions in the United States.
