Objective:

To evaluate the association of AI-generated risk scores from prior-year screening mammograms with biopsy outcomes, specifically focusing on breast cancer characteristics such as tumor grade and receptor status.

Key Findings:

• AI models demonstrated improved performance in breast cancer risk prediction compared to traditional models, indicating a potential shift in clinical practice.
• AI-generated risk scores correlated with biopsy-confirmed outcomes and clinical-pathologic features, suggesting a more nuanced understanding of breast cancer risk.
• Understanding the relationship between AI risk scores and tumor characteristics enhances clinical trust in AI systems, potentially leading to better patient outcomes.

Interpretation:

The study highlights the potential of AI in improving breast cancer risk assessment and understanding tumor biology, addressing the limitations of traditional models.

Limitations:

• The study is retrospective and may be subject to selection bias, which could affect the generalizability of the findings.
• The 'black box' nature of AI systems limits the interpretability of risk predictions, posing challenges for clinical application.
• The findings may not be generalizable to all populations due to the multicenter design, which may introduce variability in practice.

Conclusion:

AI-generated risk scores can enhance breast cancer risk assessment and provide insights into tumor biology, potentially improving clinical decision-making by offering more accurate risk stratification.