Behind the Brain

Researchers in Harvard have recently collaborated with Google to make the most complex, detailed 3D models of different brain structures. They also made these models, dubbed the Neuroglancer, available to the public here. Now, how did they make these models?

This story starts 10 years ago in the lab of Dr. Lichtman, a Harvard professor for molecular and cellular biology. In this lab, he and his team were collecting data from a cubic millimeter of brain tissue from an epileptic patient. 

Collecting the Data

They started by cutting the brain into slices using a knife with a diamond blade edge. Next, they cut them into even thinner slices, till the width of a single slice was about 0.1% the width of a hair strand. They then stained these virtually invisible slices of brain tissues with heavy metals. Lastly, through electron imaging, they obtained 300 million images of the brain tissue at an molecular level.

Data Analysis

After the procedure described above, the researchers collected 1400 terabytes of data. To put this into context, to store all the data Dr. Lichtman and his team collected, 22400 iPhones would be necessary. Hence Dr. Lichtman reached out to Viren Jain, a senior staff research scientist at Google, asking him if Google would be interested in collaborating with Harvard to analyze the 1400 terabytes of data that Dr. Lichtman and his team had collected. Why Viren Jain in particular? Well, Viren Jain was already working on the Fruit Fly project at Google. Since the Fruit Fly project was dedicated to mapping out the full brain of a fruit fly, Viren Jain already had the hardware to analyze Dr. Lichtman’s data.

To analyze the data, Google used AI-based processing and 3D imaging for two purposes. Firstly, to identify the types of cells in each image, and secondly, how they were connected. Once this model was created, they used colors to make the model easier to understand. For example, they colored inhibitory axons blue, and made excitatory axons green as shown below.

Interesting Findings & Applications

This set of 3D models, AKA Neuroglancer has several applications. Researchers are already investigating neurodegenerative diseases such as Alzheimer’s and Parkinson’s, as well as investigating memory formation. Furthermore, researchers all across the world have discovered interesting findings within the data that question our understanding of neuroscience. One such example is the unexplainable number of connections between a pair of neurons, with some neurons having over 50 connections. Another example is the discovery of axon whorls, where axons curl on top of another cell.