Hackers Teach Computers To Tell Healthy And Sick Brain Cells Apart

Mar 31, 2015
Originally published on April 8, 2015 11:17 am

Brain researchers are joining forces with computer hackers to tackle a big challenge in neuroscience: teaching computers how to tell a healthy neuron from a sick one.

"Sick neurons have a withered appearance, much like a sick plant has a withered appearance," says Jane Roskams, an executive director at the Allen Institute for Brain Science. But at the moment, she says, highly trained scientists are still better than computers at assessing a neuron just by looking at its shape, which resembles a tree that can have thousands of branches.

Automating the analysis of single neurons could greatly speed up the process, allowing analysis of thousands of cells. A standardized, computer-based system also would make it easier for researchers to compare results and allow more labs to study how the shape of neurons is changed by everything from learning to Alzheimer's disease, Roskams says.

So the institute has launched BigNeuron, a collaborative effort to improve the computer algorithms that turn microscope images of a neuron into a three-dimensional digital model and then analyze its shape. The effort will include a series of hackathons in which programmers and brain scientists get together to test their algorithms.

"So we have 15 to 20 people in a room," Roskams says. "They each have their pet algorithm, and they're kind of racing each other." The first hackathon took place in Beijing in mid-March. Others are planned for the U.S. and the U.K.

At each event, participants are given access to supercomputers and high-quality images of many different kinds of neurons. The goal is to find the best algorithms. And those won't necessarily come from people who know a lot about the brain, Roskams says.

"We have incredibly talented young people who can code and program and begin to give meaning to some of the pictures that we've been taking in a way that many neuroscientists can't imagine doing," she says.

The algorithms that emerge will be shared with scientists and even students around the world. Giving more people the ability to study neurons could help answer fundamental questions, like how the shape of a neuron changes throughout a person's lifetime, Roskams says.

"We should be able to look within an aging brain and go, "Wow, that's why that person is so sharp and sprightly. Their neuron in this part of their brain looks exactly the same as a 20-year-old's," Roskams says.

Today, analyzing the complex shape of a neuron often requires a supercomputer. But one long-term goal of BigNeuron, Roskams says, is to create programs that a high school student could use on a laptop computer.

Copyright 2015 NPR. To see more, visit http://www.npr.org/.

Transcript

ROBERT SIEGEL, HOST:

Brain researchers are joining forces with computer programmers to tackle a big challenge in neuroscience. NPR's Jon Hamilton reports the goal is to teach computers how to tell healthy brain cells from those damaged by Alzheimer's or Parkinson's.

JON HAMILTON, BYLINE: A healthy neuron looks like a healthy tree. It has a strong trunk and lots of branches. But a brain disease can change all that.

JANE ROSKAMS: Sick neurons have a withered appearance, much like a sick plant has a withered appearance.

HAMILTON: Jane Roskams is with the Allen Institute for Brain Science in Seattle. She says highly-trained scientists with high-tech microscopes can answer lots of questions about a neuron just by studying its shape.

ROSKAMS: Is a neuron healthy? What kind of a neuron is it? Is a neuron sick? What is wrong with it?

HAMILTON: In theory, a computer could answer these questions more quickly and reliably and repeat the process on many thousands of neurons. But programmers are still struggling to make that happen, so the Allen Institute is leading an effort called BigNeuron to improve the way a neuron's complex three-dimensional shape is captured and then analyzed by computer programs. Roskam says the effort includes a series of hackathons where programmers and brain scientists get together for a marathon session of testing computer algorithms that assess neurons.

ROSKAMS: So we have 15 to 20 people in a room. They've each got their pet algorithm, and they're kind of racing each other.

HAMILTON: The first hackathon took place in Beijing two weeks ago. Others will be held in the U.S. and the U.K. At each event, participants are given access to supercomputers and high-quality images of many different kinds of neurons. The goal is to find the best algorithms. And Roskams says these could come from people who know more about math and pattern recognition than they do about brains.

ROSKAMS: We have incredibly talented young people who can code and program and begin to give meaning to some of the pictures that we've been taking in the 3-D environment in a way that many neuroscientists can't even begin to imagine doing.

HAMILTON: The algorithms that emerge will be shared with scientists and even students around the world. Roskams says giving more people the ability to study neurons could help answer fundamental questions, like how the shape of a neuron changes throughout a person's lifetime.

ROSKAMS: We should be able to then look within an aging brain and go, wow, that's why that person is so sharp and sprightly. Their neuron in this part of their brain looks exactly the same as a 20-year-old.

HAMILTON: Roskams says today, analyzing the complex shape of a neuron often requires a supercomputer. But she says one long-term goal of BigNeuron is to create programs that a high school student could use on a laptop computer. Jon Hamilton, NPR News. Transcript provided by NPR, Copyright NPR.