Machine Learning Could Detect Cancer More Quickly and Accurately

Deep learning—one of the most promising approaches to artificial intelligence—could change the way cancer is detected.

The National Cancer Institute provided computer scientists with 2,000 low-dose CT scans of patients’ torsos with the goal of improving the detection of lung cancer. While the imaging technique uses less radiation than other methods, when doctors interpret the images, they tend to make false positive diagnoses—they’re not easy images for human eyes to decipher.

A CT scan of a lung tumor

It’s no secret that humans routinely miss things in images, and particularly medical images. Four years ago, researchers asked 24 radiologists to pick out a nodule on a CT scan. They correctly identified the nodule 55% of the time, but 85% of the time, they missed the giant cartoon gorilla the researchers photoshopped into the image.

Can you spot the gorilla?

Enter deep learning. The technique has become adept at classifying images and recognizing patterns in recent years. In many cases, it’s just as good as humans, and in others, it’s better. The $1 million challenge, sponsored by the Laura and John Arnold Foundation, didn’t directly pit algorithms against humans, but the results suggest that radiologists may soon have a new assistant. Here’s Will Knight, reporting for Technology Review:

The winning team employed a neural network and put extra effort into annotating images to provide more data points. It also used an additional data set, and broke the challenge into two parts: identifying nodules and then diagnosing cancer. It isn’t yet clear how the best algorithm might measure up to a doctor, because each algorithm provides a probability rather than a definitive outcome.

“We think that explicitly dividing this problem into two stages is critical, which seems also to be what human experts would do,” says Zhe Li, a member of the winning team and a student at Tsinghua University, one of China’s foremost academic institutes.

Currently, there’s no roadmap for taking the winning algorithm and turning it into a piece of software usable by doctors. But it does suggest ways in which AI could help make medicine quicker and more accurate.