Google is developing artificial intelligence to help doctors identify breast cancer, according to a research paper published Wednesday in the journal Nature,media outlet The Verge reported. According to the New York Times, the model can scan X-ray photos, known as mammograms, to reduce false negative rates by 9.4 percent, a promising leap forward for the 20 percent breast cancer tests currently missing.
Today, breast cancer is the second most common cause of cancer death in women (its mortality rate and overall incidence are second only to lung cancer). Early detection is the best defense for most people to identify and treat the disease. However, while mammograms are the most common testing tool, they miss a lot. “Mammograms are very effective, but there is still a significant problem with false negatives and false positives,” said Shravya Shetty, a Google researcher and co-author of the study. “
In the Google-funded study, researchers used anonymous mammograms of more than 25,000 women in the UK and more than 3,000 women in the United States. “We’re trying to follow the same principles that radiologists might follow,” Shetty said. According to Google’s blog post, the team first trained AI to scan X-ray images and then looked for signs of breast cancer by identifying breast changes in 28,000 women. They then examined the computer’s guesses based on the woman’s actual medical condition. In the end, they were able to reduce the false negative rate for American women by 9.4 percent and the false positive rate by 5.7 percent. In the UK, the model, which is usually examined by two radiologists, reduced the false negative rate by 2.7 per cent and the false positive rate by 1.2 per cent.
However, the system is not perfect. While researchers have found that in most cases, aI performs better than doctors in identifying breast cancer, in some cases doctors mark cancers that the model initially missed. “Sometimes, all six readers in the U.S. have found a cancer that exceeds AI, and vice versa,” Mozziyar Etemadi, a researcher at Northwestern University and another co-author of the paper, told the Wall Street Journal. “
But Google still hopes the system will eventually be used in a clinical setting. Daniel Tse, Google’s product manager, said: “We are very excited and encouraged by these results. He told The Verge that the team was working to ensure that the results of the study could be replicated among the population. He added: “When you use it for clinical practice, there are obviously a lot of nuances. “
Google has been careful to design the project to help radiologists rather than replace them. “Each of them has their own strength, which is complementary,” Shetty said. “In many cases, radiologists will find what the model is missing, and vice versa. Combining the two can enhance the overall effect. “
The project is part of Google’s expanding healthcare effort. Last year, the tech giant partnered with Ascension to gain the health records of millions of U.S. citizens. The project came under fire after a whistleblower accused health records of not being anonymous. In breast cancer research, Google, in collaboration with clinical researchers in the U.S. and the U.K., used identified data.