Clinical Scorecard: Key Insights on AI Utilization in Breast Imaging: Guidelines from the European Society of Breast Imaging

At a Glance

Key Highlights

• AI systems improve diagnostic accuracy and efficiency in mammography screening with moderate evidence supporting their use.
• Commercial AI tools for digital breast tomosynthesis and ultrasound aid lesion classification and reduce interpretation time, though clinical impact requires further evaluation.
• Robust governance and continuous post-market surveillance are essential for safe and effective AI implementation in breast imaging.

Guideline-Based Recommendations

Diagnosis

• Use AI as an aid to radiologists in breast imaging interpretation without replacing clinical judgment.
• Accept or overrule AI findings based on comprehensive clinical information without mandatory explanation in reports.

Management

• Implement AI tools following robust evidence and WHO phased evaluation, emphasizing post-market surveillance (phase 4).
• Consider AI integration in screening workflows to reduce radiologist workload and improve cancer detection rates.

Monitoring & Follow-up

• Conduct rigorous post-market surveillance of AI performance to ensure safety, quality, and efficiency.
• Monitor AI impact on long-term clinical outcomes, cost-effectiveness, and workflow changes.

Risks

• Potential variability in AI performance across different imaging modalities and populations.
• Lack of evidence for improved long-term clinical outcomes necessitates cautious adoption.
• Risk of overreliance on AI without sufficient radiologist oversight.

Patient & Prescribing Data

Women undergoing breast cancer screening and diagnostic imaging

AI-assisted mammography screening can increase cancer detection rates by 5–13% with minimal increase in recall rates; AI reduces radiologist workload significantly while maintaining diagnostic accuracy.

Clinical Best Practices

• Integrate AI as a supportive tool rather than a standalone diagnostic method.
• Ensure AI tools undergo large-scale validation and reach WHO phase 4 with continuous performance monitoring before strong clinical recommendations.
• Use decision-referral models to triage cases, optimizing radiologist workload and diagnostic accuracy.
• Maintain radiologist authority to accept or overrule AI findings based on full clinical context.

References

• World Health Organization AI Medical Device Evidence Phases