Hi Boaz,

   There are a couple of things I can think of that might be helpful.

1) Axon Identification.   Circular structures that are unmyelinated axons, as opposed to processes from oligodendrocytes, etc., should have a very distinctive array of microtubules compared to other kinds of objects in the field.  If you can add some immunolabeling to the TEM prep you could positively identify the axons using markers like Acetylated alpha tubulin or Tau.  Post-embedding immunolabeling would be optimal since you already have specimens embedded, but I don't know off the top of my head whether those antibodies have been used in that way. They do work on frozen and paraffin sections, so pre-embedding immunolabeling approaches should be feasible if that becomes necessary.

    I have never had to do this type of quantitation myself but there are a couple things I found that might be useful for image segmentation and analysis of g-ratio and comparing myelinated:unmyelinated axon ratio:

2) g-ratio.

Stikov, N., Campbell, J. S. W., Stroh, T., Lavelée, M., Frey, S., Novek, J., … Pike, G. B. (2015). Quantitative analysis of the myelin g-ratio from electron microscopy images of the macaque corpus callosum. Data in Brief, 4, 368–373. http://doi.org/10.1016/j.dib.2015.05.019

3)  For automated axon counting someone has just put together a public domain bit of software for this task. Again, I have never used it, so I can't say much about how well it works.  Zarei, K., Scheetz, T. E., Christopher, M., Miller, K., Hedberg-Buenz, A., Tandon, A., … Abràmoff, M. D. (2016). Automated Axon Counting in Rodent Optic Nerve Sections with AxonJ. Scientific Reports, 6, 26559. http://doi.org/10.1038/srep26559

Good Luck.

Dave Sherry