I have grayscale images of nuclei and their mitochondria (separate images) and I would like to measure how fragmented the mitochondrial network is; what’s the best way to do that?
Good day Sarah,
you may have a look at another thread of the Forum where I’ve provided some help: https://forum.image.sc/t/choosing-filters-for-mitochondrial-segmentation-using-trainable-weka-segmentation/20635/18?u=herbie
Perhaps the whole thread will provide some insights for planning and more detailed questions.
BTW, just a few weeks after this thread terminated I spoke to Walter Neupert (whom you surely know, at least from the literature) about mitchondria and their ability to rather quickly create networks and to dissolve them again: An interesting field of research because it seems that the reason for this is still widely unclear.
Thank you, I will read through that thread; his pictures look similar to mine.
I’m attempting to use a cellprofiler pipeline, though as I’m struggling with it I hoped there may be another approach using imageJ.
In any case we need a much more detailed description of what you like to obtain. We need to tell machines what to do and these machines understand mathematical formulations only.
to measure how fragmented the mitochondrial network is
This is not sufficient for those who are not familiar with the images in question.
Thanks for the images and thanks for making them accessible in (their hopefully original) 16bit TIF-format.
The quality of both images appears being really good.
If you like to get a global assessment of the let’s say diffuseness/fuzziness of the structures I’d recommend to start with a simple analysis of the shape (not the exact numerical values) of the image histograms.
Just take some of your images that show various degrees of diffuseness/fuzziness and look at their histograms which of course requires that you understand what they mean, but you will easily find out yourself.
If this doesn’t satisfy you, we shall go one step further and try more complicated approaches.
For the time being it’s your turn now.
Here is a more advanced but not necessarily better (when compared to the image histogram) feature function created from the two provided sample images:
Image 1 (no network):
Image 2 (diffuse network):
Thank you! I’ll give that a go. It’s nice to try something simple!
If you do want to continue in CellProfiler, you may find our Speckle Counting example pipeline and our Advanced Segmentation tutorial helpful to get going- both give examples of how to find smaller objects and measure them with respect to the larger one (the latter specifically for mitochondria and cells). You could use EnhanceOrSuppressFeatures “speckle enhancement” to find your isolated mitochondria and then just regular segmentation to find the rest.
I think there are two issues that prevent getting a sensible result:
- the resolution of the image is low to see mitochondria clearly.
- you are imaging a 3D space collapsed into a 2D image. Even if mitochondria are spatially separated the projection will show them as “merged”.
Good day Gabriel,
you are of course right in both respects, however it appears possible to use basic statistical measures to differentiate between various states of mitochodrial fusion (network building).
Whether the differentiation is sufficient, depends on the chosen measures, the images and the demands of the OP.
We shall have to wait how the OP comments our suggestions.