Finding Your Nerve

Like a having a go at ‘Where’s Wally?’, pinpointing cells in electron microscopy (EM) images of tissues is painstaking. However, it’s necessary to create 3D tissue models — this is called segmentation. Any cell running through the tissue must be identified on each slice before images can be aligned and stacked to form a 3D model. This laborious process is now automated using computer algorithms called neural networks. However, current segmentation only works well on high-resolution EM images which are time-consuming to capture. Low-resolution EM images are quicker to capture but cell borders aren’t as clearly defined. Researchers present a new segmentation method using neural networks, DeepACSON. DeepACSON was trained on high-resolution EM images of rat brains and then applied to low-resolution images of injured rat brains. It accurately stitched together 3D models of nerve axons (pictured), revealing tiny ultrastructural changes in response to injury more efficiently than previous methods.

Written by Lux Fatimathas

Lux Fatimathas

Writer

Having pursued science through a career in research, communications and publishing, she is currently a writer and editor for a healthcare marketing agency and also juggles freelance science writing.

• Image from work by Ali Abdollahzadeh and colleagues
• A. I. Virtanen Institute for Molecular Sciences, University of Eastern Finland, Kuopio, Finland
• Image originally published with a Creative Commons Attribution 4.0 International (CC BY 4.0)
• Published in Communications Biology, February 2021