An artificial intelligence program developed by Google Health has demonstrated superior accuracy in detecting breast cancer from mammograms compared to expert radiologists. In a study involving nearly 30,000 mammograms from UK and US women, the AI system identified cancers that radiologists missed while reducing false positives.
The AI was trained on mammograms from over 91,000 women and tested on a separate set. In the US, where mammograms are typically read by a single radiologist, the AI reduced false positives by 5.7% and false negatives by 9.4%. In the UK, where two radiologists review each scan, the AI marginally improved performance, cutting false positives by 1.2% and false negatives by 2.7%.
Dominic King, UK lead at Google Health, said the technology could augment human expertise. The Royal College of Radiologists noted a shortfall of at least 1,104 radiologists in the UK, with 8% of breast radiology posts unfilled. Dr Caroline Rubin, vice-president for clinical radiology, called the AI a potential "second pair of eyes" but stressed the need for rigorous testing and regulation.
The findings, published in Nature, used images predominantly from one manufacturer, and researchers caution that performance may vary with different equipment. Clinical trials are the next step to assess the AI in real-world conditions. Michelle Mitchell, chief executive of Cancer Research UK, described the research as promising but early-stage, highlighting the potential to improve screening and ease NHS pressures.
