Clinical Report: AI-Enhanced 2-Year Breast Cancer Risk Prediction in Screening

Overview

This study evaluated three machine learning models to predict 2-year breast cancer risk in women undergoing mammographic screening. The hybrid model combining clinical data and mammographic image analysis demonstrated improved risk prediction over models using either data source alone.

Background

Mammographic screening is a key tool for early breast cancer detection and mortality reduction, traditionally interpreted by radiologists. Artificial Intelligence (AI) has enhanced diagnostic performance and reduced radiologist workload, but its application for short-term breast cancer risk prediction remains limited. Breast density and clinical factors such as family history and age are established risk predictors, and integrating these with AI-extracted mammographic features may improve personalized risk assessment. Short-term risk models align with typical European screening intervals and can guide targeted interventions.

Data Highlights

The study included women aged 50–69 screened between 2013 and 2020 in a Spanish population-based program, using full-field digital mammography. Cases were women with negative mammograms who developed breast cancer within two years; controls were matched disease-free women. A 5-fold stratified cross-validation approach was used to train and test three models: (1) clinical and conventional mammographic features, (2) CNN-based mammogram analysis, and (3) a hybrid combining both data sources.

Key Findings

• The hybrid model integrating clinical risk factors and CNN-extracted mammographic features provided superior 2-year breast cancer risk prediction compared to models using either data source alone.
• AI-based mammogram analysis alone showed promising predictive capability, highlighting the value of automated image feature extraction.
• Traditional clinical risk factors and mammographic features remain important contributors to risk assessment.
• The study supports the feasibility of combining heterogeneous data sources for personalized short-term breast cancer risk prediction in screening populations.
• Use of a large, well-characterized screening cohort with rigorous cross-validation strengthens the validity of findings.

Clinical Implications

Incorporating AI-driven mammographic image analysis with clinical risk factors can enhance short-term breast cancer risk stratification in screening programs. This approach may enable more personalized screening intervals and targeted diagnostic follow-up for high-risk women, potentially improving early detection and resource allocation. Integration into clinical workflows requires further validation but holds promise for advancing precision screening.

Conclusion

Combining clinical data with AI-extracted mammographic features improves 2-year breast cancer risk prediction in screening populations, supporting the development of personalized, AI-assisted screening strategies.

References

1. Instituto de Salud Carlos III FEDER (PI17/00047) -- Study Funding and Ethics
2. European Guidelines for Breast Cancer Screening -- Screening Interval and Protocols