Setting a threshold to analyse particles

Hello everyone,

I am a new user of ImageJ, so my questions might be very basic. I use ImageJ to analyse the distribution of particles in a matrix. I have some optical microscope pictures like this one (particles are in black) :

The aim for me is to get the size distribution of the particles thanks to the Analyze particles tool.

I have some issues with setting my threshold. I was using Image > Adjust > Threshold and changing manually the upper value (picture below) :
manual threshold

While I was looking on forums for a method to choose a correct threshold value, I ended up on many topics in which it was adviced to use an auto threshold method instead. Therefore, my questions are :

  • Is using Auto threshold the most recommanded method for this kind of use?
  • Or is there any specific method to follow to choose a correct upper threshold value for my images?

Thank you in advance for any piece of advice.

1 Like

Hi there!
If you use a multi-color LUT prior to the thresholding the LUT will give you a better reference to go by.

The end goal of any analysis is to get accurate results, so you should generally go with a method that does this. Usually by checking against ground truth. If you want to dig into it, take a subsection of several images where you manually count/highlight the areas you think are positive. Test a variety of automated thresholds on those images, and compare the areas that you manually checked. The main benefit to automation is that you can’t simply choose the threshold that gives you the answer that you want.

There are many thresholding methods exactly because there is no best method for any particular project. Some will overestimate when you have very few objects, others will underestimate when you have too many positive objects… and so on.

You seem to have issues with blur around the edges (lower left and lower right) which could impact your analysis as well.

Might also consider Weka pixel classification, as it might be more robust than a threshold, depending on the consistency of your images. Making sure you are taking “fair” images is always the best place to start if you are troubleshooting.

An example of Weka in action:

Thank you for this advice, I will try to do that

1 Like

Thank you very much for this answer

1 Like