AI Breast Cancer Screening System Outperforms Human Radiologists in Study
A new artificial intelligence system surpasses human experts in predicting breast cancer from mammograms, significantly reducing both false positives and false negatives.
Researchers present an artificial intelligence system that surpasses human experts in breast cancer prediction during screening mammography. The AI model outperforms all six human radiologists in an independent study, achieving an area under the receiver operating characteristic curve that exceeds the average radiologist by an absolute margin of 11.5%.
The system demonstrates impressive error reduction by decreasing false positives by 5.7% and 1.2% in US and UK datasets respectively, while cutting false negatives by 9.4% and 2.7%. Furthermore, the AI shows strong ability to generalize its predictive capabilities across different populations and healthcare systems when transitioning from UK data to US data.
In a simulated UK double-reading process, the AI system maintains non-inferior performance while reducing the workload of the second human reader by 88%. This robust evaluation provides strong evidence that AI integration into clinical workflows paves the way for future clinical trials aimed at improving the accuracy and efficiency of breast cancer screening worldwide.