Using Ilastik to Count Clumped Cells

Hi everyone,
I’m attempting to come up with a way to automate my cell counting procedure. I’ve used several different programs, such as FIJI and CellProfiler, however I think that Ilastik is the most promising. I’ve attached some example images of what I’m working with: the cell borders are variable and sometimes pretty ill-defined when the cells are clumped. I’ve tried using the Cell Density workflow on Ilastik, but the density of cells in my images can vary a lot, and so this didn’t work consistently. I think that using the Pixel Classification followed by Object Classification (Probability Map) is the most promising, but I’m having a lot of difficulty actually getting it to work. Any advice or help would be very much appreciatedC1-Ctl03-1.tif|attachment (2.0 MB) C1-Ctl03-12.tif (2.0 MB) C1-Ctl03-4.tif (2.0 MB) C1-Ctl03-10.tif (2.0 MB)

Hi @Philippe_D_Onofrio and welcome to the forum!

I’ve had a look at all three of your images. The first one ( C1-Ctl03-12.tif looks doable with pixel classification and object classification. There you could do just foreground-background in pixel classification and then have a sort of object count classifier in object classification, with classes for false detections, single objects, two objects…

The other two, however, are just too crowded I would say. A few questions though:

  • Do I assume correctly that cells are not only touching in the “x-y” plane, but also could be on top of each other?
  • Can you rely on the size of the cells being always the same?
  • Can you by any chance add Dapi staining?

Hey K-dominik,
thanks for your reply. In answer to your questions, the cells could be on top of each other, but it shouldn’t happen too frequently. They’re always approximately the same size, and I think that this would be a good method to separate them. I can also add DAPI staining, that is not a problem at all. Knowing this, do you think there’s a way to proceed using Ilastik?

Dear @Philippe_D_Onofrio,

have you already tried:

In case it is promising, this post might be of interest to you:

Kind regards