192 — Rapid and Semi-Automated Extraction of Neuronal Cell Bodies and Nuclei from Electron Microscopy Image Stacks

Holcomb et al (10.1007/978-1-4939-3615-1_16)

Read on 28 February 2018
#segmentation  #EM  #neuroscience  #neural-network  #cell-body 

It is very useful — for a variety of reasons — to be able to locate cell bodies and nuclei in a sample of electron microscopy neuroimagery. One use-case for this information is coregistration (lining data up to functional or light-microscopy datasets); another is analysis of cell distribution in the tissue (a function of brain area and cortical depth).

Rather than segmenting the full volume of tissue (which is error-prone, time-consuming, and often extremely GPU-hungry), the authors present a simple strategy for identifying the locations of cells bodies and their nuclei.

This system uses Ilastik to drive the segmentation, and feeds the results of the data into CellSeeker, a custom-built tool that produces EM subvolumes for each cell.

This system — notably — doesn’t necessarily require neural networks (the paper is after all from 2016, before the fall of man) and this means that it’s very much faster to use the CellSeeker technique than to wait for dense deep-net-based reconstructions to render.