Silicon Valley

The firm’s deep-learning tool was able to correctly distinguish metastatic cancer 99% of the time, a greater accuracy rate than human pathologists.

The system: The team trained an algorithm (named Lymph Node Assistant, or LYNA) to spot the features of tumors that have metastasized (that is, spread), which are notoriously difficult to detect. Of the half a million deaths worldwide caused by breast cancer, 90% are due to metastasis.

Gold standard: The 99% rate is superior to the performance of human pathologists, and the algorithm was also better at finding small metastases on individual slides. Human pathologists can miss these as much as 62% of the time when under time pressure, studies have shown.

A useful sidekick: Rather than replacing humans, this technology is more likely to complement their skills, making it easier and quicker to diagnose metastatic tumors. In one study, the algorithm halved the time it took to check a slide on average, cutting it to just one minute per slide.

What next: Google is applying AI to health care in a number of different projects. Its subsidiary DeepMind is using AI to look for signs of eye disease at Moorfields Eye Hospital in London, for example. The next step will to be to see how LYNA performs in the clinic for real.