# Particle analysis of highly clustered aggregates

Dear Community,

For my research I need to analyze several pictures regarding the particle size. Unfortunately, the particles are highly aggregated making the classical “analyze particles” function very complicated.When looking at the microscopic pictures, one has the gut feeling that sample A consists of a higher number of small particles compared to sample B. This gut feeling I need to express in some kind of measure.
So far I played around with different commands and filters including among others treshhold, sharpen, watershed but I do not find a proper combination of these that would give me a reasonable number so to speak.
Do you have any idea how to proceed on that?Sample A.tif (3.0 MB) Sample B.tif (3.0 MB)

I’m grateful for every answer. Cheers!

@Paco

This is a difficult task indeed. Looking at your images… it’s pretty much impossible to delineate individual particles. If it’s difficult for our eyes - it will be difficult for a computer. So to this end… I’d say you have two choices: 1) come up with another way to mathematically define a difference between these two conditions (overall area, roughness measure, etc.) or 2) try to work on the sample prep side of things to get more isolated particles. I’m sorry I cannot be more help in this… perhaps others have ideas as well?

2 Likes

Thank you for your answer. Roughness did not really work but it brought me to the idea of texture analysis with the GLCM Texture plugin. I found it hard to get proper information on how the plugin actually works and how I can interpret the results.
According to this lecture (http://www.cyto.purdue.edu/cdroms/micro2/content/education/wirth06.pdf), GLCM “contains information about the positions of pixels having similar gray levels”. My idea now was that in sample A, the particles are smaller and therefore pixels with similar gray levels are more distributed. Hence, the entropy in sample A ist higher. And the results look more or less like this.
This is really just a rookie assumption based on my admittedly little knowledge concerning imageJ. But do you think I might be onto something here.

Best regards

1 Like

@Paco

Sounds promising to me thus far! For sure… this type of analysis based on your images (and I’m assuming limited sample prep options) is more appropriate. If you have more questions… you can always start a new thread. We are all here to help!

1 Like

Hello Paco,
I am not sure what it is you are trying to measure, however an analysis of just the histogram comparisons may be all you need.
Bob

2 Likes

@smith_robertj
I’ll be looking into that, too, in a while.
Thanks for the input!