Stand-Alone Artificial Intelligence for Breast Cancer Detection in Mammography: Comparison With 101 Radiologists

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Oncology

Background:

• Digital mammography (DM) is crucial for breast cancer screening.
• AI systems have the potential to enhance DM interpretation accuracy and efficiency.
• Evaluating AI performance against radiologists is key for clinical adoption.

Purpose of the Study:

• To compare the standalone performance of an AI system with radiologists in detecting breast cancer using DM.
• To assess if AI can achieve radiologist-like accuracy in DM analysis.

Main Methods:

• Utilized nine multi-reader, multi-case datasets from seven countries, totaling 2652 DM exams (653 malignant).
• Included interpretations from 101 radiologists (28,296 total interpretations) and an AI system's analysis.
• Compared AI and average radiologist performance using area under the ROC curve (AUC) and a noninferiority test.

Main Results:

• The AI system's performance was statistically noninferior to the average of 101 radiologists.
• AI achieved an AUC of 0.840 (95% CI: 0.820-0.860), compared to 0.814 (95% CI: 0.787-0.841) for the average radiologist.
• The AI system outperformed 61.4% of individual radiologists.

Conclusions:

• The AI system demonstrated comparable cancer detection accuracy to an average breast radiologist in this retrospective study.
• The findings are promising for AI's role in breast cancer screening.
• Further investigation into AI performance and impact in actual screening settings is warranted.