How to Discriminate Between Activated and Not Activated Neurons in Fluorescence Staining

Hi everyone!

I’m a new user of ImageJ, and I would like to discriminate activated neurons versus non-activated neurons after a stimulus based on the fluorescence intensity of those neurons.
I stained brain tissue sections (25µm thick) with antibodies that targeted the marker of neuronal activation c-Fos and revealed activated neurons in red thanks to a red fluorophore attached to the antibody.

However, after a long optimisation of the staining protocol, I have difficulties in determining which neurons are activated.
Indeed, I have different light intensities for activated neurons on a single image!
It may be caused by the image acquisition (but I can’t modify it), by the depth of the tissue ( but I can’t make thinner sections) or by a bad staining (but I already tried a lot of conditions).

(Another point is that my DAPI staining didn’t work properly, so I can’t detect the neurons via an independent channel: I’ll have to detect AND measure neurons with one staining only, so automated threshold will create a bias on neurons selection…)

So the critical point is to know which threshold to apply so I can be sure that above this threshold, cells are activated.

Pre-processing does not help to chose the right threshold ( I tried to subtract the background, to apply a Gaussian blur, …).

Do you think a statistical determination of the threshold by the plugin Statistical Region Merging could work? What about using hysteresis thresholding?

I attache a sample picture to show you my problem:

Thank you in advance for your help.

@stelfrich, do you have any clue? :smile:

Good day Baptiste,

as far as I understand, the main question must be answered by yourself.

If you can’t tell which intensities mean inactivity and which do not, you wont be able to set a proper theshold.

Not sure someone will be able to help you with this evidently underdetermined problem.



Thank you very much for your answer Herbie,

Indeed, I don’t know the exact intensity that means activity.
But, what is certain is the fact that the cells I study have only two possibilities: being activated or not. Consequently, we can suppose that an intensity stage exists (between activity or inactivity).
So I hope that somehow, with a statistical method based on an analysis of both populations of neurons (activated or not), I could estimate the stage of activity. (I think that this method is close to hierarchical clustering)

Another option would be to determine the intensity of cells I consider as activated (because they represent the maxima of activity), then estimate the intensity of others activated neurons based on this sample (another problem emerging is how to choose this sample).

Does it seem feasible/useful to you?
Best regards

Dear @Baptiste,

Well, you seem pretty certain. However, I doubt that this assumption holds.

Generally speaking, there are a lot of options. As @anon96376101 has pointed out, you will first have to define what you actually want to measure: are you interested in the threshold value itself (if so, how will you interpret it subsequently?) or in the percentage of cells in an active state vs non-active state?

Judging from your sample image, I’d say that it will be pretty hard (if possible) to establish a robust, automated pipeline for segmentation.



Well Baptiste,

as you may expect, I second Stefan and I highly recommend to re-think your task or at least formulate it more stringent.

My academic mentor used to say:
“If you can mathematically define what is signal and what is noise, you are nearly done.”

You know best how your marker reacts to neural activity – whích by the way?

What does the granularity all over the image mean? Is it an artifact on the slide?

Why is it that “DAPI staining didn’t work properly”?

Too many questions yet.



Dear @stelfrich, thank you so much for your fast answer,

So, I looked back at the theory, and indeed, you’re right. Because the cFos fluorescence staining method is by itself continuous (and not discrete), we can’t categorize neurons activities. Consequently, neurons that have a low intensity are considered as having less activity than the ones having a high intensity: we can’t know if the differences in intensities are biologically meaningful, so we have to decide empirically how to process them.

Knowing what I said before, I’m now interested in binarizing my set of images so that I consider the low intensities always the same way (this is now the critical point), no matter how I consider them.

1 Like

Your questions helped me a lot in re-thinking my task! Thank you very much @anon96376101 and @stelfrich!

The granularity is caused by fluorescence staining: it’s due to non-specific fixations of the reagents. (I guess it follows a Gaussian repartition)

DAPI staining doesn’t show any cells under the microscope, this is just an observation

1 Like