AI software as a third reader in breast cancer screening-a prospective diagnostic observational study.

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Oncology

Background:

• Mammography screening is crucial for early breast cancer detection.
• Despite advancements, some cancers remain undetected, necessitating improved screening methods.
• Artificial intelligence (AI) is being explored to enhance mammography's diagnostic accuracy.

Purpose of the Study:

• To evaluate the effectiveness of AI software (Transpara) as an independent third reader in mammography screening.
• To determine if AI integration can reduce missed cancer diagnoses.
• To assess the impact of AI on cancer detection rates and positive predictive values.

Main Methods:

• A prospective study enrolled 15,356 women in German Mammography Screening.
• Mammograms underwent double reading and independent analysis by Transpara AI software.
• Consensus conferences reviewed cases flagged by readers or AI; endpoints included cancer detection rate (CDR) and positive predictive values (PPV).

Main Results:

• AI as a third reader increased the cancer detection rate by 9.5% compared to double reading.
• The AI primarily detected Luminal-A-like cancers; nine invasive cancers were detected solely by AI.
• The positive predictive value for consensus-referred cases decreased with AI integration, and the number of recalled cases increased.

Conclusions:

• Integrating AI as a third reader improves mammographic cancer detection, particularly for Luminal-A-like cancers.
• AI offers complementary sensitivity but may miss aggressive subtypes like triple-negative breast cancers.
• Human readers remain essential, and AI integration necessitates careful consideration of increased workload and recall rates.