Google's AI May Replace Doctors In Detecting Breast Cancer Because It's More Accurate

Google has just reported success in training AI to detect breast cancer more accurately than doctors do. In particular, a Google-funded study gathered independent researchers from different universities and hospitals, Google Health researchers and DeepMind engineers to compare analyses of almost 29,000 mammograms from datasets collected in the US and the UK.

The team found that the US had a 9.7% reduction in false negatives, 5.7% in false positives while the UK had a 2.7% decline in false negatives as opposed to 1.2% in false positives. AI achieved these results although it was given less information. In its control study, the company randomly selected 500 mammograms and gave them to six radiologists, along with the patients’ previous mammograms, breast cancer history, and their age.

Google doesn’t provide all the demographics information of its survey sample, but if the AI works with a more diverse sample that represents the real world, it can help detect cancer that’s hidden behind the human tissue. It’s also indicated in the paper that this could help radiologists significantly reduce their workload.

The new decade will surely bring more leaps with Google expanding into the field of healthcare. According to Google CEO Sundar Pichai, healthcare is one of the biggest fields of artificial intelligence where it takes between 10 and 20 years to see the benefits.

The latest study is the company’s new attempt in detecting the spread of breast cancer using deep learning. In the last few years, the tech giant has been training AI to detect heart disease and diabetic eye disease, as well as analyzing data on multiple sclerosis’ progression. Back in 2014, Google attempted to create a smart contact lens embedded with a microchip to detect levels of blood sugar in tears, but the project was canceled because that wasn’t enough as a data source.

However, professor Etta Pisano, a respected radiologist from the Beth Isreal Lahey Medical Center, warned the medical community from being too excited about the technology. He pointed out that detection relying on the computer may fail, such as CAD, first unveiled in the 90s and showing great potentials in experimental testing, failed in the real world. The technology did not diagnose more accurately than humans.

According to Pisano, CAD tended to over-mark images of shaped that didn’t turn out to be real cancer. But he also writes that screening for breast cancer can be a promising application for artificial intelligence regarding medical imaging since there’s a huge dataset for AI to work with as well as a more binary determination compared with diagnoses that involve manifold factors.

Let’s hope that Google will dig deep and come up with something really useful and accurate in the coming years.