While LLMs have been in the spotlight lately, different types of AI techniques are being introduced across the sciences.For example, researchers at Harvard University, Google, and his colleagues are using 3D maps to science This week, a tiny part of the human brain is revealed in astonishing detail. Imaging approximately cubic millimeters of tissue generated 1.4 petabytes of data.
nature has a great summary article He then quotes Biren Jain, a neuroscientist at Google in Mountain View, California, and co-author of the paper: How can we actually embrace all this complexity? ” Jain's team then built an artificial intelligence model that could stitch together the microscopic images and reconstruct the entire sample in his 3D. “I remember this moment when I walked into the map and looked at her one individual synapse from this woman's brain, and then zoomed out to millions of other pixels.” says Jane. “It felt kind of spiritual.”
The work itself has been going on for about 10 years. “About 10 years ago, a small piece of a human brain arrived in the lab of Dr. Jeffrey Lichtman at Harvard University. According to a report in the Boston Globe, it was shipped directly from a nearby hospital's operating room. It was said to have been removed from an epileptic patient undergoing treatment to alleviate seizures there.
A single neuron (white) and its 5,600 axons (blue) are shown. The synapses that make these connections are shown in green. Credit: Google Research & Lichtman Lab (Harvard University). Rendering by D. Berger (Harvard University)
“Over the next years, Lichtman's team systematically reconstructed the brain's Byzantine wiring patterns by feeding 1 cubic millimeter samples into a $6 million machine and slicing them incredibly thin. We then used electron microscopy images of those fragments to painstakingly reconstruct the complex lattice structure that interconnects each individual cell.”
The image is remarkable. The summary represents the scale of the project.
“Presented here is a computationally advanced ultrastructural reconstruction of a cubic millimeter of human temporal cortex that was surgically removed to access the underlying epileptic focus. It contains approximately 57,000 cells, approximately 230 millimeters of blood vessels, and approximately 150 million synapses, with an area of 1.4 petabytes. 2:1, oligodendrocytes are the most common cells, and deep excitatory neurons can be classified based on the orientation of their dendrites, with thousands of weak connections to each neuron. We showed that there are rare and powerful axonal inputs within up to 50 synapses. Further studies using this resource may yield valuable insights into the mysteries of the human brain. ”
The full paper is available in Science, but Google and the researchers provided the following information: public access Connect to the dataset and Neurglancer viewer to view your data.
Link to scientific paper (petavoxel fragment of human cerebral cortex reconstructed at nanoscale resolution), https://www.science.org/doi/10.1126/science.adk4858
Link to Boston Globe article written by Adam Piore https://www.bostonglobe.com/2024/05/10/metro/google-harvard-brain-map-images/
Link to article in Nature, https://www.nature.com/articles/d41586-024-01387-9
related
