To characterize and compare specific mammographic features in screening examinations with high AI risk scores for women with and without screen-detected breast cancer.
Key Findings:
739 screen-detected cancers and 208 interval cancers were identified, with implications for clinical practice.
High AI risk scores were associated with both positive and negative cases, indicating potential for false positives and the need for careful interpretation.
Mammographic features identified by AI models differed between women with and without cancer, suggesting areas for further research.
Interpretation:
AI models can identify suspicious mammographic features, but their application may lead to false positives, necessitating careful interpretation and follow-up, particularly in training AI models.
Limitations:
Retrospective design may introduce bias, potentially affecting the reliability of findings.
Exclusion of certain cases could limit generalizability, impacting the applicability of results.
Dependence on AI model performance and accuracy may influence the outcomes and their interpretation.
Conclusion:
AI-assisted screening has the potential to enhance mammographic interpretation but requires further validation in practical settings to optimize recall rates and reduce false positives.
by Marit A. Martiniussen, Marie B. Bergan, Merete U. Kristiansen, Nataliia Moshina, Anne Sofie F. Larsen, Marthe Larsen, Fredrik A. Dahl, Solveig Hofvind
