Infrastructure automation for connectomic image analysis

点击翻译按钮获取中文摘要

The BRAIN 2025 report states that a major goal of the US BRAIN Initiative is "Generate circuit diagrams," and identifies electron microscopy (EM) as "the gold standard for circuit mapping." So far EM is the only approach that has ever delivered a connectome, a map of all synaptic connections in a nervous system or brain. After the C. elegans connectome in the 1980s, the labor of manual image analysis prevented the EM approach from generalizing to larger nervous systems. Since then, labor has been dramatically reduced by progress in artificial intelligence. Humans need only correct the errors that remain in an automated reconstruction. Zetta AI was founded to make connectomic image analysis accessible to any neuroscientist. In 2021, Zetta completed an automated reconstruction of a cubic millimeter cortical volume for the Allen Institute. This is one of only three existing petascale reconstructions in the world. For the Harvard Medical School, Zetta also completed an automated reconstruction of the Drosophila ventral nerve cord. These successes establish Zetta as a leading organization in connectomics. Zetta’s image analysis pipeline requires significant engineering labor to operate. Based on our operations over the past two years, we have identified several opportunities for engineering labor reduction by process automation, including EM image ingestion, image alignment, and hard example mining. Such process automation will help make connectomics accessible to all neuroscientists. Availability of neural circuit diagrams will aid the discovery of connectopathies and other structural pathologies that have long been hypothesized to be associated with brain disorders.