Normalization of fluorescence images

measure
thresholding
Tags: #<Tag:0x00007fd54872ab80> #<Tag:0x00007fd54872a4a0>

#1

I am new to ImageJ and I am seeking a bit of clarification about how to quantify my fluorescence images. I have performed an experiment with two treatments and I want to quantify the fluorescence intensity for both treatments to compare them. Here is a simplified version of what I did:

  1. Image > Adjust > Threshold > Auto > Set
  2. Analyze > Set Measurements > “check” limit to threshold
  3. Analyze > Measure

I then compared the mean that was returned for each treatment. However, I noticed that the means returned for my images relatively equal even though there was a visible difference in fluorescence intensity on the actual raw images. I also noticed that the MinThr and MaxThr for the images were very different and I decided to perform min max normalization between 0-1 to make the means more comparable.

Example:
Image 1
Mean= 840.529
MinThr=423
MaxThr=1708

Image 2
Mean=812.471
MinThr=492
MaxThr=2437

Is it correct to perform the min max normalization??


#2

Hello Olivia, can you post the two example images you mention?


#3

Hi Svere,
I have posted examples of Image 1 and Image 2:


#4

First of all, I am sorry for responding and then not answering again untill now! I have been very busy and kind of forgot this post but; I had some time today.

So I must admit I don’t quite understand how you did this, it might be the easier way to do it. Also I think you should compare each treatment to control, rather than directly to each other (image of the cells before and after treatment, for each treatment, and then compare the relative effect). But this is how I would have done it for the analysis anyhow:

I don’t like measuring the whole image and taking the mean after thresholding for reasons. So what I did as simple as possible, is the following:

Note: This workflow might not work for full automation if you have a lot of data. You would probably have to apply more sophisticated segmentation techniques.

  1. Apply gaussian blur.

  2. Threshold to create a mask, make sure I get all the cells. Do you want to also measure the nucleus? It seems your tag localizes to the cytosol. If you do you can just adjust the threshold or jo can measure the nucleus and cytosol individually, see region 6 and region 5 below. but whole cells is probably fine for you.

  3. Apply irregular features watershed plugin to separate neighboring cells. This way we get individual measurements for each cell.
    plugins > BioVoxxel > Irregular Features Watershed > play around with settings, look at their wiki.

  4. Edit > selection > create selection (on the mask) > ctrl+t (add to manager)

  5. Delete artefact ROIs, this is more elegantly done in a script if you decide to automate this, with a macro or script (if size below 800 or above 3000 delete ROI).

  6. Apply selection to raw image and measure.

And then you are free to choose how to display your data. You can do a mean of all the cells or mean*area. I don’t know what is preferred. Personally I prefer to analyse individual cells. Maybe someone else has input on this.

This is what I get when I plot sum of mean intensity of all cells in each image, where series 1 is image one and series 2 is image 2.