Best way to segment synaptic vesicles on EM images


I’m facing a bit of a challenge here. I am trying to segment synaptic vesicles on electron microscopy images of synapses. Basically, they look like small, round vesicles (dark ring, see image 1), but depending on conditions they can also be larger, with a more irregular shape (see image 2). The active zone of the synapse (where vesicles fuse) is the darker interface between the cells, at the bottom of Image 1 and top of Image 2.


Image 2:

What do you think would be the best tool for this? Classical machine learning or deep-learning? I’m more used to the Fiji ecosystem (so that would be classic Analyze particles, Weka segmentation, StarDist plugin…) but I can learn new things too. If someone has time to tackle the challenge and show me how their favorite tool can do it, I’m all eyes and ears. Thanks for the help!

Hi @christlet,

I don’t know if it’s the best way, but with CellPose I got the following ROIs:

It is far from being perfect and it missed a few vesicles but possibly with some optimisation it could be improved.

I used @pr4deepr’s notebook that you can find here.
I made first a Gaussian blur (s=1) then run CellPose with object diameter ranging from 7 to 25 and combined the outputs.

Some vesicles appear as one :point_down:
But it’s just because of the way I combined all the outputs. In the actual masks they are separated :point_down:.

Hope this helps.


That’s pretty impressive for a largely automated result!

1 Like