A new study published in Radiology reveals that artificial intelligence (AI) models developed during the 2023 RSNA Screening Mammography Breast Cancer Detection AI Challenge can effectively identify different cancers during screening mammography.
Led by Professor Yan Chen, PhD, of the University of Nottingham, researchers evaluated over 1,500 AI algorithms submitted to the global competition. The study assessed how well the models detected cancer, and whether combining top-performing algorithms—known as ensemble models—could improve diagnostic accuracy.
Key Findings
■ Among 1537 AI algorithms evaluated in the RSNA Screening Mammography Breast Cancer Detection AI Challenge, the median recall rate, sensitivity, and specificity were 1.7%, 27.6%, and 98.7%, respectively.
■ Combining the top 3 or top 10 algorithms increased the sensitivity to 60.7% and 67.8% and recall rates to 2.4% and 3.5% while decreasing specificity to 98.8% and 97.8%, respectively.
■ The sensitivity of the top 3 ensemble model was greater for invasive than for non-invasive cancers (68.0% vs 43.8%; P = .001).The study analysed data from over 5,400 women across the U.S. and Australia, using pathology-confirmed cancer cases and rigorous follow-up for non-cancers.
“These findings demonstrate the potential of AI to enhance early breast routine screening,” said Prof. Chen. “Ensemble models, in particular, can increase diagnostic sensitivity without increasing unnecessary recalls.”
This work underscores the exciting future of AI in breast screening, while highlighting the ongoing need for robust validation in diverse clinical populations.
Full article here https://pubs.rsna.org/doi/10.1148/radiol.241447
