Artificial Intelligence Could Help Spot Cancer
Neural networks are learning to identify cancer cells in images of tumor tissue
Deep-learning technology—a form of artificial intelligence that trains algorithms to recognize patterns in massive datasets, such as a set of images—can be used to identify cancer cells in microscopic snapshots of tumor tissue, Heather Couture, a computer scientist at the University of North Carolina, Chapel Hill, writes in a commentary in Scientific American.
In order to train typical artificial neural networks to spot objects such as cars, trees, and even faces, computer scientists organize vast data sets, and deep learning algorithms look for similarities within the data. But spotting cancerous cells has its limitations because large data sets containing images of these cells are scarce. To overcome this challenge, Couture and her team relied on a method called “deep transfer learning,” in which neural networks adapt their process to recognize more generic features instead of finer details.
The algorithm was also able to spot the difference between some cancer subtypes, including more aggressive ones. Couture and her team hope this new method will save time and money, and most importantly, lives.