Potential Impact of an Artificial Intelligence-based Mammography Triage Algorithm on Performance and Workload in a...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Healthcare

Background:

• Screening mammography is crucial for early breast cancer detection.
• Current mammography workflows face challenges with performance metrics and workload demands.
• Artificial intelligence (AI) offers potential solutions for optimizing screening processes.

Purpose of the Study:

• To evaluate the performance and workload impact of a commercial AI-based triage device in a population-based screening mammography sample.
• To compare AI triage performance against radiologist reports and actual outcomes.
• To assess the theoretical workload reduction achievable with AI triage.

Main Methods:

• A retrospective study analyzed 2129 screening mammograms.
• A commercial AI triage device categorized cases as 'suspicious' or 'low suspicion'.
• AI performance was compared with radiologist reports, actual outcomes, and national benchmarks using key mammography metrics, with up to 5 years of follow-up data.

Main Results:

• At 93% sensitivity, AI triage significantly improved recall rate (45.5%), PPV1 (119%), PPV2 (28.5%), sensitivity (20%), and specificity (7.2%).
• A theoretical workload reduction of 62.5% was observed at 93% sensitivity and 27% at 99% sensitivity.
• No cancers were missed by the AI algorithm at either tested sensitivity setting.

Conclusions:

• AI-based triage shows significant potential to enhance mammography performance metrics.
• The AI tool can lead to substantial theoretical workload reductions in screening mammography.
• The implementation of AI triage in this simulation did not compromise cancer detection rates.