FIJI for particle size distribution of crystals in pharmaceutical formulations

Dear All,

I would like to do particle size distribution for crystals found in pharmaceutical formulations. So far, I have learned a lot about converting images to 8-bit, about setting the scale (I have a scale bar in all my images), about thresholding (manual or auto), about making binary images and about using the watershed option for separating crystals that are in contact…and, in the end - about the option “analyze particles”, and its possibilities…

All of this works well for simple cases - like in the example no.1, where only small particles are present, all in the same focal plane…

example1.tif (14.1 MB)

The problem are images with more particles, some of which are out-of-focus (these particles should not be measured!)…watershed option also did not work for this image…

example.tif (14.1 MB)

Please advise me on what would be the right way to do this image.

I believe any pipeline that you recommend will be better then the present situation in which the operator sits in front of the computer and manually draws lines (the major diameter) on each and every one of the sharp (focused) particles, which, of course takes forever, and is extremelly subjective and error-prone…not to mention that it drives people insane (and blind), since several visual fields must be measured (for example, 500 particles are measured in an average analysis)…did I mention that it takes forever? :slight_smile:

Thank you very much in advance…


Hi Jelena,

this is a REALLY difficult problem - I once had a project where we were trying to count ice crystals in images that looked relatively similar to your second example, and we never got a satisfactory result. I even presented in the Call4Help section at the last NEUBIAS conference and asked the audience for suggestions. The presentation is here and the google doc with the suggestions is here (I was the second presentation on this document).

Unfortunately, even with the best minds put together in the same room, none of the suggestions worked for my problem. That shouldn’t discourage you; that document has a list of things you can try and that might work! At the end, the closest thing to a reasonable solution we’ve found was a graph-based segmentation algorithm. I have a github repository with my (very poorly documented) code here. Hope any of this helps!

Hi Eric,

thanks for your reply. Unfortunately, I was unable to solve my problems after I went through everything you have sent. It seems that my best chance is to work on better sample preparation, in order to get only one layer of crystals, all in focus…but, that seems imposible at the moment :frowning:

Please…anybody else?