On May 10, 2024, AI techniques beyond just LLMs are gaining prominence across scientific endeavors. A notable example comes from Harvard researchers, Google, and their collaborators, who unveiled a highly detailed 3D map of a small portion of the human brain in a recent publication in Science. The imaging of approximately one cubic millimeter of brain tissue resulted in a staggering 1.4 petabytes of data.
Nature provides a concise overview of this achievement, quoting Viren Jain, a neuroscientist at Google, who remarks on the complexity of the endeavor, stating, "It’s a little bit humbling. How are we ever going to really come to terms with all this complexity?" Jain's team employed artificial intelligence models to merge microscope images and reconstruct the entire sample in 3D. Jain reflects on the profound experience of exploring the intricate details, describing it as "sort of spiritual."
The project itself has been in progress for about a decade. It began when a small piece of human brain tissue was delivered to Dr. Jeffrey Lichtman's lab at Harvard from a nearby hospital's operating room, where it was excised from an epilepsy patient undergoing treatment to alleviate seizures.
Over the years, Lichtman's team meticulously reconstructed
the brain's intricate wiring patterns using advanced techniques. They sliced
the 1-cubic-millimeter sample into extremely thin sections and then utilized
electron microscopy to capture images of these slices, piecing together the
complex network connecting individual cells.
The magnitude of the project is evident from the abstract, which outlines the reconstruction of a cubic millimeter of human temporal cortex. This sample contains approximately 57,000 cells, 230 millimeters of blood vessels, and 150 million synapses, totaling 1.4 petabytes of data. The analysis reveals intriguing findings, such as the prevalence of glia cells outnumbering neurons, the classification of deep layer excitatory neurons based on dendritic orientation, and the existence of rare powerful axonal inputs.
The full paper is accessible through Science, with Google
and the researchers offering public access to the dataset and Neurglancer
viewer for further exploration.
Science Paper: A
petavoxel fragment of human cerebral cortex reconstructed at nanoscale
resolution
Boston
Globe Article by Adam Piore: Google, Harvard unveil brain map images
Nature Article: Mapping
the human brain with advanced AI techniques