127 — 3D Axonal Morphometry of White Matter

The long axons that extend great distances through the mammalian brain are often encased in myelin sheaths, which increase the propagation speed of action potentials. Much of the existing electron-microscopy segmentation and analysis algorithms that operate on these long, myelinated axons operate in 2D space, leveraging the image-stack nature of the EM data.

ACSON, the pipeline described in this paper, segments largescale ultrastructure in serial block-face scanning electron microscopy (SBEM). While the segmentation algorithm used (bounded volume growing, or BVG) doesn’t quite work for grey-matter segmentation challenges, it achieved Dice coefficients in the high eighties and low nineties for myelinated axons.

Abdollahzadeh et al then compared metrics of an induced TBI rat and a control — a sham-surgery rat brain — based upon this segmentation. The researchers found that the cross-sections of the axons along their primary axes were more often elliptical than circular, and rats that had undergone TBI exhibited axonal swelling, with more damage occurring in lightly myelinated axons than in heavily myelinated ones.

I would be interested to see how this pipeline performs if the segmentation algorithm were replaced with another more robust one, from some of the papers I’ve discussed before. I imagine even more subtle morphological patterns would become evident with a more sensitive segmentation algorithm.