Advanced deep learning models are revolutionizing long-term breast cancer risk prediction by analyzing serial screening images over time. Traditional risk models mostly rely on clinical factors or single-timepoint 2D mammograms. However, sequential digital breast tomosynthesis screening provides richer temporal and volumetric data for clinicians.
Longitudinal AI in Breast Cancer Risk Prediction
Researchers developed a deep learning model to estimate two- to five-year breast cancer risk using sequential screening tomosynthesis scans. Specifically, the study evaluated over 300,000 screening examinations across a diverse cohort of 161,077 women. The longitudinal algorithm combined temporal image features, patient age, and volumetric breast density parameters. Consequently, the dynamic artificial intelligence system captured progressive anatomical changes that single-timepoint images often miss.
Comparing Risk Prediction Models
In an independent test set of 34,570 women, the longitudinal model achieved a five-year AUC of 0.721. Notably, this performance significantly surpassed single-timepoint tomosynthesis algorithms and full-field digital mammography models. Furthermore, the new algorithm markedly outperformed traditional clinical tools like the Tyrer-Cuzick risk model. In addition, the deep learning tool successfully reclassified high-risk patients who lacked obvious traditional clinical risk factors.
Clinical Implications for Screening Programs
Therefore, tracking sequential tomosynthesis scans allows radiologists to establish dynamic risk profiles for individual patients. Additionally, personalized risk stratification helps clinicians tailor supplemental screening strategies, such as contrast-enhanced mammography or MRI. As a result, early intervention strategies can target high-risk individuals before mammographically visible lesions appear.
Frequently Asked Questions
Q1: What is longitudinal digital breast tomosynthesis?
Longitudinal tomosynthesis involves evaluating a patient’s sequential 3D mammograms across multiple years rather than analyzing a single scan in isolation.
Q2: How does the new model improve breast cancer risk prediction?
The model detects subtle progressive tissue changes over time, offering significantly higher predictive accuracy than traditional single-timepoint clinical calculators.
Q3: Can AI models replace clinical risk tools like Tyrer-Cuzick?
AI models complement clinical risk tools by providing precise image-based risk scores that capture dynamic structural changes in breast tissue.
References
- Xu Y et al. Predicting 5-Year Breast Cancer Risk From Longitudinal Digital Breast Tomosynthesis: A Single-Center Retrospective Study. AJR Am J Roentgenol. 2026 Aug 12. doi: 10.2214/AJR.26.34951. PMID: 42584410.
- Lehman CD et al. Longitudinal Analysis of Changes in Deep Learning Image-based Breast Cancer Risk Scores over Time. Radiology. 2026;310(3):e231840.
- Jiang S et al. Deriving a mammogram-based risk score from screening digital breast tomosynthesis for 5-year breast cancer risk prediction. Cancer Prev Res. 2025;18(6):345-353.