How AI Helps Create a Detailed High-Resolution 3D Map of the Human Brain


Published on:

Key Takeaways:
  • Using a combination of high-speed electron microscopy and machine learning, researchers have created the most detailed 3D map of the human brain ever.
  • The detailed brain map is freely available on the Neuroglancer platform, allowing researchers worldwide to explore the human cortex at an unprecedented level of detail.

Creating a detailed atlas of the human brain has long been a goal for neuroscientists. A collaboration between Harvard scientists and Google has brought us a step closer to achieving this ambition. Together, they have developed a nanoscale 3D map of just one cubic millimeter of the human brain. While this represents only a tiny fraction of the entire brain, which is a million times larger, this small segment is densely packed with approximately 57,000 cells, 230 millimeters of blood vessels, and around 150 million synapses. This makes it the most detailed image of the human brain available so far.

To produce such a detailed map, the research team sliced a tissue sample into 5,000 thin sections and scanned each using a high-speed electron microscope. Afterwards, they employed a machine learning model to digitally reconstruct and categorize these slices. The sheer amount of raw data accumulated to 1.4 petabytes. “It’s probably the most computer-intensive work in all of neuroscience,” noted Michael Hawrylycz, a neuroscientist at the Allen Institute for Brain Science, who was not involved in the project. “There is a Herculean amount of work involved.”

Unlike earlier brain atlases, which feature much lower resolution, this nanoscale map allows researchers to meticulously trace each neuron and its connections down to the synapses — the crucial points where neurons connect. “To really understand how the human brain works, how it processes information, how it stores memories, we will ultimately need a map that’s at that resolution,” explained Viren Jain, a senior researcher at Google and co-author of the study published in the journal Science. The dataset was first released in 2021.

Brain atlases are available in various forms. Some illustrate the organization of cells, while others explore gene expression, particularly focusing on the connections between cells — a field known as “connectomics.” The outermost layer of the brain alone contains about 16 billion neurons, which are linked through trillions of connections. A single neuron can both receive input from hundreds or thousands of others and send information to a similar number of neurons. This complexity makes the task of tracing these connections highly intricate, even when it involves only a small portion of the brain.

To create the map, the researchers faced challenges obtaining suitable brain tissue. Cadaveric tissue was not feasible since it deteriorates quickly. They used a sample from a woman undergoing brain surgery for epilepsy. The fresh tissue provided the viable cells needed for high-resolution mapping.

The researchers preserved the obtained sample with resin and sliced it into extremely thin sections — about a thousandth of the thickness of a human hair — and then imaged using a high-speed electron microscope that was developed specifically for this project.

The process of reconstructing the three-dimensional connections of the wires in the brain slices was a significant computational challenge for the Google team. Viren Jain explained that it requires applying machine learning techniques to align and color code the wires to identify their connections accurately. This process is more complex than it may seem as a single error can lead to mistakes in all subsequent connections.

“The ability to get this deep a reconstruction of any human brain sample is an important advance,” states Seth Ament, a neuroscientist at the University of Maryland. He describes the map as “the closest to the ground truth that we can get right now.” However, Ament also highlights a limitation — the map is derived from a single brain sample from one individual.

The data of the human cortex is now available on a web platform called Neuroglancer to assist researchers in their studies. The platform offers a map that provides detailed information that can be used to verify specific structures and ensure accuracy before publishing findings. According to Viren Jain, this level of detail offers anyone interested in exploring the human cortex the opportunity to examine the data firsthand.

The research team has stumbled upon a few unexpected discoveries. For instance, the nerve fibers that look like tubes and transmit signals between neurons have formed “vortices” or areas where they twist around themselves. Normally, these axons create a single synapse to pass information to the next neuron. However, the team found individual axons that formed multiple connections, sometimes up to 50 separate synapses. The reason behind this is still unclear, but these robust connections might facilitate very rapid or intense responses to certain stimuli, as suggested by Jain. He notes that, “It’s a very simple finding about the organization of the human cortex. But we didn’t know this before because we didn’t have maps at this resolution.”

Jeff Lichtman, a neuroscientist at Harvard University who co-led the study, affirms that the data set is filled with revelations. “There were just so many things in it that were incompatible with what you would read in a textbook.”

Related Posts:

Leave a Reply

Please enter your comment!
Please enter your name here