Overview
- Two Nature Cancer papers report AI flagged about 25% of interval cancers on earlier mammograms and identified more invasive cancers than a single human reader.
- Using AI as the second reader cut estimated screening reads by roughly 32–40% while maintaining double-reading standards in large NHS analyses.
- Real-world feasibility work in London showed faster AI reads (about 18 minutes versus over two days for first human review) but higher arbitration rates and trust frictions, including 93 AI-correct cases overruled.
- A separate npj Digital Medicine study of 112,621 negative screens found MIT’s Mirai best in head-to-head risk prediction, capturing 27.5% of interval cancers within the top 4% high‑risk group and showing some sensitivity to imaging hardware.
- An NHS Grampian trial using Mia on 10,889 women yielded 11 additional cancers (seven invasive), reduced unnecessary recalls, shortened notification time to three days, and reported up to 31% workload savings.