It may come as a surprise that although we know much about how the eye works, neuroscience researchers do not understand how visual signals translate into perception.
We’ve landed on Mars, can grow organs, and even skydive from space, yet when it comes to a thorough understanding of the territory so close to home that it is home, much is missing. Neuroscientists don’t even know precisely how many different types of cells are in the brain. Here at Sebastian Seung‘s Computational Neuroscience Lab at MIT, we’re taking a different approach: crowd-sourcing. In order to solve the mind’s great mysteries, we need you.
Why don’t we know how the mind works? One reason is that your mind is massive. Researchers estimate that there are 100 billion neurons in your brain with an estimated one million miles of connectivity [1,2]. A million miles is equivalent to driving around Earth 40 times. You can infer that in order for such great length of neurons to fit into your three pound brain these structures must be very tiny. A large neuron is about 100 microns in diameter while the contact area of a synapse is less than 7 cubic nm [3, 4: data from Seung Lab].
In order to see neurons and the tiny structures called dendrites through which they function, researchers utilize a new imaging technique. “Fix whole brain tissue, slice off layers just a few microns thick, image each slice with an electron microscope, and trace the path of each neuron,” explains David Zhou, Masters Student at Carnegie Mellon, on Quora. These gamechanging techniques generate a terabyte of data for even a cubic millimeter of brain tissue. Now that we can see the brain at the synaptic scale, we have to analyze the images. How?
The image above shows the process of layering image slices to render 3D reconstructions. Like most neuroscience labs, the Seung Lab uses a combination of AI algorithms and tracing (3D reconstruction) performed by humans. Why not just use algorithms? Images can be challenging to identify, particularly for a computer. Pure algorithms make many mistakes, such as slicing a single cell into thousands of pieces and merging multiple cells into one monstrously massive neuron. See below image for an example of AI missing a chunk of a neuron.
We hope to one day train computers to map neurons on their own; however, that day will be far in the future and we need to accelerate neuroscience discovery now. To achieve this, we need something more intelligent than even the most powerful supercomputer— you.
It takes an MIT-trained neuroscientist anywhere from 15 to 80 hours to reconstruct a single neuron. At that rate, it would take about 570,000,000 years to map the connectivity of an entire human brain, known as a connectome. This is why we need your help.
By playing the 3D game Eyewire, you become part of the Seung Lab at MIT by helping to map the connections of a neural network.
Rather than mapping and entire brain, we’re starting with a retina. Our goal is to map the connections of a specific type of cell: J-Cells. These neurons are responsible for perception of upward motion. We plan to publish the outcome in a scientific journal and list EyeWire users as co-authors.
Scientific American writes that “no specialized knowledge of neuroscience is required [to play EyeWire]; citizen scientists need only be curious, intelligent and observant. Your input will help scientists understand how the retina functions. It will also be used by engineers to improve the underlying computational technology, eventually making it powerful enough to detect “miswirings” of the brain that are hypothesized to underlie disorders like autism and schizophrenia.”
Here’s a short video from Sebastian with more info: