Connectomics from electron microscopy

Reconstructing neural connectivity from electron microscopy requires imaging at synaptic resolution across volumes that can reach terabytes, together with reliable alignment of thousands of tiles per section and hundreds of sections in depth. In this line of work we contributed methods and tools that span the stack: automatic mosaicking and slice-to-slice registration (ir-tools), web-scale volume browsing and annotation (Viking), axon tracking in serial block-face SEM, hierarchical and cascaded models for EM image segmentation, and joint ultrastructural and molecular imaging frameworks that combine serial-section TEM with computational molecular phenotyping. Applications included the RC1 retinal connectome dataset and workflows for whole-section brain mosaics at light-microscopy scale. This project is retained as a reference for prior EM connectomics contributions; it is not an active research thrust in the current group portfolio.

Together these efforts aimed to make connectome-scale EM analysis more practical—reducing manual labor for alignment and tracing, and providing classifiers that respect the noise and anisotropy of real EM stacks—while keeping human annotation in the loop where automation is still insufficient.