Researchers showed that Alphabet’s (GOOG, GOOGL) DeepMind AI unit could match or outperform radiologists at detecting breast cancer, though doctors still beat the machines in some cases.
Google has announced that it has trained an AI model that improves the screening of breast cancer (detection before symptoms develop). After two years of collaboration with DeepMind, Cancer Research UK Imperial Centre, Northwestern University, and Royal Surrey County Hospital, the resultant AI model “spotted breast cancer in de-identified screening mammograms (where identifiable information has been removed) with greater accuracy, fewer false positives, and fewer false negatives than experts.”
The model was trained on anonymized, de-identified mammograms from more than 76,000 women in the United Kingdom and more than 15,000 women in the United States. The trained model was then evaluated by a de-identified dataset of more than 25,000 women in the United Kindom and over 3,000 women in the United States. This led to the model producing impressive results:
In this evaluation, our system produced a 5.7 percent reduction of false positives in the U.S, and a 1.2 percent reduction in the U.K. It produced a 9.4 percent reduction in false negatives in the U.S., and a 2.7 percent reduction in the U.K.
Moreover, in a separate experiment, a dataset of women in the United Kingdom was used to train a model that was later validated on the dataset of women in the United States. This was done to investigate whether the model could be generalized to other healthcare systems. Yet again, the model reported increased accuracy in the screening process:
…a 3.5 percent reduction in false positives and an 8.1 percent reduction in false negatives, showing the model’s potential to generalize to new clinical settings while still performing at a higher level than experts.
These are impressive results, considering the AI demonstrated superior accuracy despite working with limited information—only the most recent anonymized, de-identified mammograms—unlike doctors who have access to patient histories and prior mammograms:
Despite working from these X-ray images alone, the model surpassed individual experts in accurately identifying breast cancer.
