Difficult Cell Counting

Hey guys
can someone help me to count these cells. i try it a couple of times but dont get it really yet …

thx

1 Like

Hi @Milo85

Are there 3 cells in this image or many small ones in the read area?

1 Like

Hi @Milo85,
I think that you can try to approach this problem with the Find Maxima… (in Process menu) after maybe some filters to enhance the difference between cells and their surrounding.
I did not try but I think that you can reach a good estimation of cells number.
Have a nice day,
Emanuele

there are many in the red are. the bigger light green ones are dirt.

the small dark red points are the cell nucleous, the length ones are the arms of the cells. important are the cells with the nucleous in it

i hope u understand what i am writing :slight_smile:

1 Like

can u perhaps explain it more detailed…

would be great.

thank u

Hi @Milo85,
unfortunately, or better fortunately ;), I will go in vacation next Thursday and I’ll be back after two weeks.
If I have time in next days I will help you for sure!
have a nice day,
Emanuele

I am not sure if I get it. Could you circle some example cells by hand?

And do you have the images in a different format (tiff)?

Take a look

i upload it here (tiff)

1 Like

Thanks for the example!

Unfortunately I couldn’t get a really good segmentation of the cell nuclei.

I tried the following:

  1. Select the red channel and invert.
  2. Apply Difference of Gaussian (DoG) (sigma1 = 2, sigma = 3)
  3. Threshold (value = 6)

The result contains some segments which are not cell nuclei but have the same shape/size. This means filtering is not an option.

1 Like

I guess you could use machine learning (e.g. the Trainable Weka Segmentation plugin) to train a classifier that enhances/segments the cell nuclei.

Have a look at this discussion for an example use case.

2 Likes

Hi @Milo85,
I have tried something like that:
0) convert your image to a grayscale image (image->type->8 or 16bit)

  1. perform a LoG filtering (you can use for example the one present in the FeatureJ plugin http://imagej.net/FeatureJ)
    in my case i tried with a sigma of 3 (but maybe other value is better) : this enanche the difference between cells and their surrounding
  2. optionally smooth more with process->filter->Median or Gaussian
  3. Process-> Find Maxima with noise tolerance of 1 this is the process that permits you to find cells.
    If you select output type : Point selection you can add them to roimanager as multipoints http://imagej.net/docs/guide/146-19.html#sec:Multi-point-Tool, then you can add or delete point pressing mouse+shift or mouse+alt.

I attach you my result: it’s not perfect of course but maybe it could be enough for your analysis.

you could also select the area of investigation to remove some borders and outside borders failures of the analyis.

To better results maybe it is the case to use some more complex techniques like Machine Learning as suggeseted by @imagejan

Have a nice day,
Emanuele

0988-1297.zip (8.0 KB)

2 Likes

it looks very good … but i cant handle it :frowning:

I know it looks overwhelming, but if you follow the steps outlined by @emartini one by one, you can do it! And if you get stuck, ask again here and we will help you! :persevere:

1 Like