Measuring fluorescence intensity on a z-stack

Image Analysis
#1

Hi everybody,

I am new to the image J forum.
I am also measuring flourescence intensity, so would like to check my protocol and ask some thinks I am not sure about.

I need to compare caspase-3 staining between two groups of cells (control and patient cell lines). Rather then comparing intensities of the individual cells, I need average intesity of all the stained cell nuclei across the image.

Here is what I am doing:

  1. I did z-stack imaging, so first I creat a z-projection using max intensity method.
  2. Then I split channels to get the channel that shows just caspase staining
  3. Duplicate the image
  4. Adjust threshold manually on duplicated image (I use fill holes and watershed option if necessary)
  5. I enabled limit to threshold in set measurements and redirected measurements to the first image
  6. Then I do measurements with analyse particles (I have chosen both mean gray value and integrated density to measure, but I think integrated density would be a preferable choice). I have checked summarize box to get mean integrated density of all the measured cell nuclei in the image.

Images are calibrated.

  1. I was using manual threshold. However I am a bit confused now. It seems that both manual and auto threshold could produce biased measurements. If I use manual threshold with always definied range would that be ok? Or is there a certain auto threshold method that wouldn’t produce biased results on my images?

Just to reflect on the problem that was mentioned in the comments: I was using the same channel for threshold and measurements, but I can see I will have to use DAPI to define threshold, and do the measurements on the caspase-3.

  1. Should I measure the intensity of the background too as they did it here:
    http://theolb.readthedocs.io/en/latest/imaging/measuring-cell-fluorescence-using-imagej.html

Here is example image:

MAX_Stack
MAX_Stack1024Ă—1024 107 KB

I would be greatful if somebody could clear this things for me :grinning:
Thanks

2 Likes
Measuring fluorescence intensity
#2

Hi @dsoltic,

welcome to the forum!

In many studies I’ve seen, the total intensity (for example in a cell) is the target measurement. But you cannot measure the total intensity (sum intensity of all pixels) from a maximum projection, as the maximum projection does not sum pixels in Z, as the name suggests. That’s why I would recommend doing a SUM projection. You may consider using a MAX projection to get a good segmentation, but later you should analyse signal intensities in the SUM projected image.

Manually adjusting a threshold is a method which has limited reproducibility. If you want to write scientific paper, it may be hard to explain how you determined the threshold. I would recommend using one of the methods listed in the pulldown. You could then write something like “Segmentation of the cells was done using the method from Otsu et Al. (1979).” It’s not that an automatic method is per definition better or worse than manual thresholding, it is just more reproducible. And what is science without reproducibility? :wink:

I agree, as “integrated density” is the thing I called “total intensity” above. But again: SUM projection, not MAX projection. Furthermore, try not to use the term “integrated density” in your scientific report. Fluorescence micrsoscopy has nothing to do with density measurements. You are measuring the “toal signal intensity” of a cell. The software just calls the thing wrong.

In general, I can recommend finding a good method for normalisation of measured signal intensities. However, I cannot say if placing an ROI close to the segmented object is per se the right method. It depends on the used imaging method, camera offset and some more parameters. At the end: You need to justify how you normalise your measurements. Ask yourself: Is the sum over all pixels in Z (SUM projection) in an area where there are no cells a good reference? Do I sum pixels which are bright because of also out-of-focus light? Is my illumination in all images the same? …

I hope that answers at least some of your questions! :wink:

Cheers,
Robert

5 Likes
Nuclear pore counting!
#3

Thanks for the help Robert,

You definitely gave me something to think about :grinning:

Cheers,
Darija

1 Like
#4

Hi Darija,

I would recommand you to have a look to this previous topics :

Cheers,

Romain

5 Likes
#5

Thanks Romain,

The question of choosing the right threshold method is more clear now.

Darija

#6

Hi @haesleinhuepf,

I am also trying to quantify fluorescence intensity in a z-stack, but I noticed that the SUM projection method in ImageJ is not a true sum of pixel intensity in the z-direction for any pixel (x,y). It actually subtracts intensity from each pixel, and results in an image that is less intense (visibly) than the MAX projection.

Do you know the reason for this subtraction?

gitHub code for ImageJ z-projection - line 343

#7

Hi @Bibi_Sulaman,

the lower visible intensity is reasonable given the sum projections characteristically different from the max projection. In case you doubt the correctness of values of a projection, I recommend using the pixel inspector to check them manually:

image
image1114Ă—650 332 KB

There you can also see that the intensity in the SUM_ image is much higher than in the MAX_ image. Which makes sense, right?

The code line you pointed at is related to 16-bit images of type unsigned integer. This subtraction is necessary to map the pixel values correctly. So it’s just a type conversion; nothing serious that can harm your analysis. If you want to be sure that this piece of code does not influence your image analysis, do the projection twice: Once with the image of type 16-bit and once again with type 32-bit. The results should be the same.

I hope that explanation helps.

Cheers,
Robert

1 Like
#8

Thank you for responding @haesleinhuepf!

I understand that the subtraction is to convert intensity values between different types of images, and ideally, no signal should be lost in that process.

However, I used the pixel inspector tool on one of my images and I’m actually getting lower intensity in the SUM projection than in the MAX projection.

Max
Max666Ă—762 274 KB
sum
sum682Ă—764 203 KB

(I care about the green channel)

The same result occurs when I measure Integrated Density in the same ROI applied to the two projections (sum < max).
intensity%20result_annotated

I am starting with RGB images, though, so could that be affecting the result? Any ideas to explain this/am I missing something?

Also, note that we expect the signal in our images to be quite faint and sparse (as you may have noticed) due to the fact that it is representing fluorescently tagged mRNA molecules at baseline in our cells.

Many thanks!

#9

Oh, that doesn’t look good. I agree. Just some thoughts:

  • You might consider processing the channels individually and not as RGB image. RGB might screw things up in this context.
  • Working with JPG files is not a good idea as its compression might introduce artefacts.
    Would it be complicated to change your workflow to multichannel TIF files? That could resolve your issues…

Cheers,
Robert

#10

Good to know about the JPG compression.
I’ll look into your suggestions - thank you!

-Bibi

1 Like
#11

Hi @haesleinhuepf,
Would it be acceptable to quantify intensity in Z-projections alone, when the scope did not save each slice of the stack? I’m under the impression we shouldn’t use Z-projections for quantification, but you answer seems to point that it’s ok to use SUM projections. Could you clarify that?
Thanks so much!
Danuta

1 Like
#12

Hey @dsastre,

nice yo hear from you again!
If you work with SUM projections, signal intensity quantification is doable. The SUM projection sums in Z, afterwards, you sum in X and Y. If you can then measure the volume (area in X/Y times first/last slice distance in Z) you can get the average intensity.

Does that answer your question?

Cheers,
Robert

#13

Thanks, @haesleinhuepf! It makes sense that SUM projections could be used. The images I’m tasked with quantifying are max contrast projections created by the scope. I’ll see if I can get the volume for those. I’m thinking in this case all I could get is a positive/negative classification with false positive cells.
Thanks,
Danuta

1 Like