Artificial Intelligence Versus Radiologist False-Positives on Digital Breast Tomosynthesis Examinations in a...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• False-positive findings in mammography artificial intelligence (AI) can inform strategies to reduce recall rates.
• Understanding AI's false-positive characteristics is crucial for its clinical integration in breast cancer screening.

Purpose of the Study:

• Compare the characteristics of false-positive digital breast tomosynthesis (DBT) examinations between AI and radiologists.
• Evaluate AI's performance against radiologists in identifying false positives during breast cancer screening.

Main Methods:

• Retrospective analysis of 3183 screening DBT examinations from 2977 women (mean age 58).
• A commercial AI tool analyzed DBT images; radiologists provided interpretations.
• False positives were defined as no breast cancer diagnosis within 1 year; radiologists re-reviewed AI-flagged findings.

Main Results:

• Both AI and radiologists had a 10% false-positive rate.
• AI-only false positives were associated with older age, fewer dense breasts, and more prior breast cancer history/procedures compared to radiologist-only false positives.
• Concordant false-positive findings between AI and radiologists had a high rate (44%) of yielding high-risk lesions upon biopsy.

Conclusions:

• Significant differences exist in patient and imaging characteristics between AI and radiologist false-positive DBT findings.
• While overlap is small, concordant false positives represent a potentially enriched subset of actionable abnormalities.
• Findings can guide AI implementation to enhance DBT recall specificity in breast cancer screening.