Google AI Breast Cancer Tool Reduces Screening Errors Across Borders
DeepMind and Google Health create an AI system that outperforms human radiologists in detecting breast cancer, even when trained on data from a different country.
DeepMind and Google Health develop a new artificial intelligence system that assists doctors in detecting breast cancer early. The researchers train the algorithm using mammogram images from female patients in the US and UK, and it outperforms human radiologists by significantly reducing both false negatives and false positives. This advancement addresses a critical issue in global women's health, as breast cancer remains the most common cancer for women worldwide and the second leading cause of cancer death.
The AI system shows impressive results in lowering screening errors across different patient populations. For US patients, the technology reduces false negatives by 9.4% and false positives by 5.7%, while UK patients see reductions of 2.7% and 1.2% respectively. In a separate experiment designed to test the model's ability to generalize, the researchers train the system exclusively on UK mammograms before evaluating it on US patients. The system still outperforms human experts, cutting false negatives by 8.1% and false positives by 3.5%.
This ability to generalize across borders offers promising implications for the future of AI in healthcare. It suggests that developers may overcome one of the largest hurdles in medical AI adoption, which is the need for massive, highly representative datasets for every single patient population. However, experts caution against overconfidence, noting that the US and UK share relatively similar demographics and the system might not perform as well in other parts of the world.