LAUSR

Biologists have long used electron microscopy (EM) to examine the nanoscale structure of biological matter. Image segmentation groups image pixels together into labeled regions corresponding to image content, and it is a fundamental tool in the quantitative analysis of EM data. Historically, segmentation was done by hand; it is a tedious procedure, but could be completed within acceptable time frames for the quantities of data produced by electron microscopes. However, modern EM hardware is capable of generating gigapixel 2D images and teravoxel 3D image volumes thanks. One technology in particular which contributes to this trend is serial block-face scanning electron microscopy (SBF-SEM) [1], which generates large 3D image volumes by scanning the face of a large sample block, shaving off a few nanometers with a diamond ultramicrotome, and repeating this process until the sample is consumed. This technique offers to radically improve our understanding of the organization of biological systems, but it is infeasible to segment large data regions by hand.