Story at a glance
- The study published in Nature shows Google's software detected breast cancer more accurately than human radiologists.
- The study found a reduction in the number of false negatives and false positives in the AI's diagnosis.
- Google says the study shows AI could enhance radiologists' capabilities in breast cancer screenings.
Google says it has developed an artificial intelligence (AI) system that appears to outperform doctors at detecting the presence of breast cancer, findings which highlight the potential for the technology to improve screening for the second leading cause of cancer fatalities among women.
A study published in the scientific journal Nature on Wednesday tested the accuracy of Google’s AI tool, which examined mammogram images from tens of thousands of women in the United States and the United Kingdom.
The study found Google’s software spotted cancer with more accuracy than human radiologists. The software’s review of mammograms resulted in a reduction in the number of false negatives, where existing cancer goes undetected, as well as in the number of false positives, where tests wrongly show signs of cancer.
Compared to human experts, the number of false negatives fell by 9.4 percent in the U.S. and by 2.7 percent in the UK, while the number of false positives dropped by 5.7 percent in the U.S. and 1.2 percent in the UK.
Google Health says researchers trained the AI system using mammograms from more than 76,000 women in the UK and more than 15,000 from the U.S. The software was more accurate despite having less information than doctors who had access to prior mammograms and patient histories.
In another test, the researchers put the AI against six radiologists presented with 500 mammograms. Overall, the software did better than doctors, but in some instances, the AI missed a cancer the radiologists found, and vice versa.
According to the American Cancer Society, clinicians fail to find about 20 percent of all breast cancers, and about half of the women getting annual mammograms over a 10-year period will have a false positive finding at some point.
The study’s authors wrote in Nature that the “performance of even the best clinicians leaves room for improvement,” and the “AI may be uniquely poised to help with this challenge.”
Google Health says the study sets the stage for “future applications where the model could potentially support radiologists performing breast cancer screenings.”
The new tool was developed by Google Health and its British subsidiary DeepMind, as well as experts at Northwestern University and radiology researchers at Cancer Research U.K. Imperial Centre and Royal Surrey County Hospital.