Related Experiment Video
Updated: Jul 9, 2026
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
42.6K
AI-supported approaches for mammography single and double reading: A controlled multireader study
Beniamino Brancato1, Veronica Magni2, Calogero Saieva3
1Unit of Breast Imaging, Istituto per lo Studio, la Prevenzione e la Rete Oncologica (ISPRO), Florence, Italy.
European Journal of Radiology
|April 22, 2025
Summary
Artificial intelligence (AI) significantly boosted radiologist sensitivity in mammography interpretation, especially for less experienced readers. AI as a second reader in double reading improved cancer detection without lowering specificity.
Area of Science:
- Radiology
- Medical Imaging
- Artificial Intelligence
Background:
- Mammography interpretation relies on radiologist expertise.
- Varying experience levels can impact diagnostic accuracy.
- Artificial intelligence (AI) offers potential for decision support.
Purpose of the Study:
- To evaluate the effect of AI on radiologist diagnostic performance in mammography.
- To compare AI's impact across different radiologist experience levels.
- To assess AI in single and simulated double reading scenarios.
Main Methods:
- Retrospective analysis of 150 mammograms (30 malignant, 120 benign).
- Five reading approaches: human single, AI single, human with AI support, human-human double, human-AI double reading.
- Sensitivity and specificity calculated and compared using statistical tests.
Main Results:
- AI-supported single reading increased mean sensitivity from 69.2% to 84.5% (p < 0.001).
- Sensitivity gains were most pronounced in lower-performing radiologists.
- Human-AI double reading achieved 91.8% sensitivity, outperforming human-human double reading (87.4%, p=0.016).
Conclusions:
- AI significantly enhances mammography sensitivity across reading paradigms.
- AI particularly benefits radiologists with lower baseline performance.
- AI as an independent second reader improves sensitivity in double reading without sacrificing specificity.
