Artificial Intelligence as Supporting Reader in Breast Screening: A Novel Workflow to Preserve Quality and Reduce...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Mammography screening is crucial for early breast cancer detection.
• Current double-reading protocols by human readers can be resource-intensive.
• Optimizing AI integration into screening workflows is an active area of research.

Purpose of the Study:

• To evaluate an AI strategy as a supporting second reader in mammography screening.
• To assess the effectiveness of AI in detecting breast cancer compared to traditional double reading.
• To determine the impact of AI support on radiologist workload.

Main Methods:

• Retrospective analysis of a large-scale, multi-site, multi-vendor dataset (280,594 mammograms).
• AI acted as a second reader only when disagreeing with the first human reader's assessment.
• Statistical analysis included non-inferiority and superiority testing of screening performance and workload reduction metrics.

Main Results:

• The AI-supported reading strategy was superior or noninferior to human double reading across all screening metrics.
• Workload reduction was substantial, decreasing the need for a second human reader by up to 87%.
• Cases requiring arbitration were significantly reduced when AI acted as a supporting reader.

Conclusions:

• The proposed AI-supported reading workflow maintains breast cancer screening performance.
• This strategy offers significant potential for reducing radiologist workload in mammography screening.
• Further investigation into the impact on second human readers is warranted.