“This approach has the potential to revolutionize biomedical research,” Margaret Sutherland, program director at the National Institute of Neurological Disorders and Stroke, which partially funded the work, says in a statement.
The researchers, who published their work in Cell today (April 12), designed their a neural network, a program modeled after the brain, using an approach called deep learning, which uses data to recognize patterns, form rules, and apply those rules to new information. “We trained the neural network by showing it two sets of matching images of the same cells; one unlabeled and one with fluorescent labels,” coauthor Eric Christiansen, a software engineer at Google Accelerated Science, says a press release. “We repeated this process millions of times. Then, when we presented the network with an unlabeled image it had never seen, it could accurately predict where the fluorescent labels belong.”
With high-quality images, the program correctly identified nuclei within cells just about perfectly. It could also distinguish dead from living cells, spot neurons within groups of cells that included astrocytes and immature dividing cells, and even tell a dendrite from an axon.
“Techniques like this tend to have a democratizing effect,” opening up opportunities for smaller groups, Molly Maleckar, the director of mathematical modeling at the Allen Institute for Cell Science who was not involved in the study, tells Wired.
The authors of the study say future work will look to optimize the network and boost its performance on certain tasks where it was less robust, such as picking out neuronal subtypes and finding axons in high-density cultures.