Google Releases Most Detailed 3D Brain Scan Image in History

2020-01-23

Google and its partners today released the largest and most detailed brain scan collection ever. The project covers nearly one-third of the fruit fly brain and includes detailed mapping of 20 million synapses and over 25,000 neurons. They chose fruit flies because they are complex enough compared to previous subjects, but still have small enough brains to make the work controllable. Unlike worms or other lower animals, fruit flies have a built-in navigation system and have advanced behavioral functions such as courtship dances. They are very similar to us, just on a much smaller scale. In neuroscience, a long-standing hypothesis is that connectivity between brain cells plays an important role in brain function. Historically, technical challenges have been obstacles for neuroscientists trying to study brain networks in detail, but this situation is beginning to change. Last year, Google announced the first nanoscale reconstruction of the entire fruit fly brain, which focused on the individual shapes of cells. However, this achievement did not reveal information about their connectivity.


Today, Google, in collaboration with HHMI's Janelia Research Campus' FlyEM team and several other research partners, released the "hemibrain" connectome, a highly detailed image of neuronal connections in the fly brain, along with a set of tools for visualization and analysis. The hemibrain comes from 3D images of about half of the fly brain and contains validated connectivity between approximately 25,000 neurons, forming over 20 million connections. So far, this is the largest brain connectome with synaptic resolution in any species. The goal of this project is to create a public resource that any scientist can use to advance their work, similar to the fruit fly genome that was released 20 years ago and has become a fundamental tool in biology.


Imaging, reconstructing, and validating the hemibrain connectome required decades of research and development. At Janelia, new methods were developed to stain fly brains and then divide the tissue into individual 20-micron-thick sections. Custom focused ion beam scanning electron microscopy was then used to image each section at 8x8x8nm3 voxel resolution, in a continuous operation lasting several months. Computational methods were developed to stitch and align the raw data into a coherent 26 trillion pixel 3D volume.


However, without precise 3D reconstruction of neurons in the fly brain, it is not possible to generate a connectome from this type of imaging data. Google established a collaboration with Janelia in 2014 to start studying fly brain data, with a focus on achieving automation of 3D reconstruction to jointly work towards building connectomes. After several iterations of technological development, a method called Flood-Filling Networks (FFN) was successfully designed and applied to reconstruct the entire hemibrain dataset. In the current project, we worked closely with collaborators to optimize the reconstruction results, making them more useful for generating connectomes (i.e., embedded in alignment and synapse detection pipelines) rather than just displaying the shape of neurons.


Finally, after validation, the reconstruction was combined with automated synapse detection to generate the hemibrain connectome. Janelia scientists manually labeled individual synapses and then trained neural network classifiers to perform the task automatically. Multiple rounds of labeling improvements were made to enhance generality, and the results from two different network architectures were merged together to generate reliable classifications throughout the entire hemibrain.


What has been released?

The focus of today's announcement is a set of interconnected datasets and tools that allow anyone interested to study fly connectomes visually and programmatically. Specifically, the following resources can be used:


  • Several terabytes of raw data, aligned 3D reconstructions, and synapse annotations that can be interactively visualized or batch downloaded.
  • The web-based tool neuPrint, which can be used to query connectivity, partners, connection strengths, and morphologies of any specified neurons.
  • Downloadable, compact representations of the connectome that are about a million times smaller in byte size compared to the raw data.
  • Documentation and video tutorials on how to use these resources.
  • A preprint with further details on the production and analysis of the hemibrain connectome.

Researchers have already started using the hemibrain connectome to gain a deeper understanding of the fruit fly nervous system. For example, a major brain circuit of interest is the "central complex".

RECENT AI NEWS

RECENT AI TOOLS