Google Says Its A.I. Can Spot Lung Cancer When Doctors Don't

Google’s artificial intelligence model could help catch lung cancer earlier, which currently ranks as the biggest cause of death in the world. The breakthrough, announced Tuesday, could raise survival rates to a staggering 40 percent.

“We know that when cases are diagnosed early, patients have a higher chance of survival, but unfortunately, over 80 percent of lung cancers are not caught early,” Lily Peng, from the Google A.I. healthcare team, said on at the Google I/O developer conference in Mountain View, California.

Google’s A.I. can help improve these rates, through a specially-trained machine learning model that looks at images and flags when an image resembles cancer. Peng explained that low-dose computed tomography scans currently used to look for lung cancer can help catch it early, but the team’s A.I. could help doctors identify cancer on these scans much earlier.

It’s the latest in a range of breakthroughs, announced at the conference, that can make machines smarter. The show was peppered with announcements about Google Assistant getting faster and Duplex taking on movie tickets. New products like the Nest Hub Max help thread these underlying advancements into consumer-facing products.

But while booking movie tickets is neat, it’s these advancements in healthcare and industry where Google’s expertise really shines. DeepMind, the London-based A.I. firm owned by Google parent company Alphabet, explained in 2016 how its systems could identify oral, neck and head cancers and save time for doctors.

Verily, another Alphabet firm, used millions of robot-raised mosquitos to help fight the Zika virus in December 2018. Calico, also part of the parent company, is looking even further to longevity and anti-aging, an area that some experts believe could lead to the first 1,000-year-old person.

Google’s Lung Cancer A.I.: How It Boosts Survival Rates

Peng explained that lung cancer “causes more death than any cancer.” It ranks as the most common cause of death globally, accounting for three percent worldwide.

When caught earlier, patients have a higher chance of survival. Stage I lung cancer at the five-year mark offers a 50 percent chance of survival, but stage IV cancer at the same mark has just a two percent survival rate. Over 80 percent of lung cancers are not caught early.

In a paper the team is about to publish in Nature Medicine, a neural network was trained on deidentified lung cancer scans from the National Cancer Institute and Northwestern University. By looking at a number of examples of malignancy, the model was able to identify cancer at a rate meeting or exceeding those from trained radiologists.

The breakthrough comes from the fact that early stage cancer is very small, but very much present, on early scans. Patients with late stage lung cancer, when it’s easier to spot, do show signs of cancer in these earlier scans.

Peng gave the example of a patient that took a routine CT scan and seemingly showed no signs of cancer. One year later, the patient showed signs of stage III cancer. Google showed the initial scans to radiologists, and five out of six were unable to spot the cancer. The A.I., on the other hand, picked out the cancer on the earlier scans. This early identification could boost survival rates to 40 percent.

“Clearly this is a promising but early result, and we are very much looking forward to partnering with the medical community to use technology like this to help improve outcomes for patients,” Peng said.

It probably won’t see as much use among the general public as an even faster voice assistant, but the machine learning tool could transform society in a much more meaningful way.