Breast Cancer Detection

AI readers of mammograms perform similarly to individual radiologists in detecting breast cancer but fall short of the accuracy achieved by multi-reader systems used in screening programs in Australia, Sweden, and the UK, highlighting the need for human-AI collaboration. Using a large dataset from Victoria, Australia, this study simulated AI-integrated screening pathways, showing that AI as a second reader or high-confidence filter can improve sensitivity by 1.9–2.5% and specificity by up to 0.6%, while reducing assessments and human reads. However, automation bias negatively affects multi-reader settings, emphasizing the need for careful design and further research before clinical implementation.