What if deep learning models could detect early-stage cancer more accurately than a veteran radiologist?
Google is showing that day might be close, based on the results of a study that is set to publish soon in the journal Nature Medicine.
“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,” said Lily Peng, a product manager at Google Brain AI Research.
Lung cancer is by far the leading cause of cancer deaths, and the risk of developing it is around 1 in 16 people.
Speaking at the Google I/O developer conference, Peng explained how the company had trained algorithms to look at CT scans in search of early signs of lung cancer.
“Very early stage cancer is minuscule and can be hard to see, even for seasoned radiologists, which means that many patients with late-stage lung cancer have subtle signs on earlier scans,” said Peng. “By looking at many [CT scans], the model learns to detect malignancy, with a performance that meets or exceeds that of trained radiologists,” she said.
In one case, the model was able to identify early signs of lung cancer in a patient who later developed the disease, even though 5 out of 6 radiologists missed those signs. Google worked with partners at the National Cancer Institute and Northwestern University on the project. Peng called the results “promising, but early.”
This is familiar turf for Google. The company has also used deep learning models to diagnose diabetic retinopathy, a leading cause of blindness, based on retinal scans. Google has also used AI to assess the risk that prostate cancer poses and to detect metastatic breast cancer.
The AI work is part of a larger health push at Google, which is also looking closely at wearables. Two years ago, the company bought health monitoring startup Senosis, a spinout of the University of Washington, and hired founder Shwetak Patel. Google’s move into this area is part of a broader push into digital health by the traditional tech industry, including other big companies such as Amazon and Microsoft.
Although the company’s Google Fit efforts have largely been overshadowed by the Apple Watch health monitoring features, Google parent company Alphabet isn’t giving up. Verily, a subsidiary of Alphabet, is working on a range of health-focused wearable projects, from smartwatches to smart contact lenses — even smart shoes.