Dense cell population cell count

Hello,

I need some help with determining the cell count and area of my ganglia. I have very little experience in doing image analysis and could really use some help.
So this is what one of my images looks like

Could anyone point me in the right direction on how to proceed with this image.
I can not get a good separation of my cells because declumping them based on intensity or shape is not working very well.
So what kind of pre-processing do you suggest?

Hello Lieke01,
I would start with the Plugin > Polynomial shading correction , with settings equal to your x,y ratio. from there you should be able to adjust brightness/contrast. That at least would give you a better start to further analysis.
Good luck,
Bob

Hi @Lieke01,

I checked your image. Hope your raw image is better than the one posted here. It is very much possible with CellProfiler too to get the cell count & area. You could look for cellprofiler tutorials available if you are a new user. In addition there are example pipelines in various level that could be downloaded along with their respective images. You could try them to start with, particularly check out fruit fly & translocation pipelines. Though the goal may not be same but with respect to cell segmentation you might check out.

In case you need further helps, let us know. In that case it would great if you share the raw image file here.

Regards,
Lakshmi
Fujifilm Wako Automation (Consultant)
www.wakoautomation.com
For CellProfiler training or optimised pipeline write to,
lakshmi.balasubramanian.contractor@fujifilm.com

Read more on our site.
Yokogawa CV8000 - The Ultimate in Confocal HCS
https://www.wakoautomation.com/products/yokogawa-high-content-imaging

Hi,

Thank you for your suggestions. I have already looked into the translocation pipeline, but I will definitely try to look into the fruit fly one.
I followed a few tutorials online, but I find it still quite challenging to get it right. One of the main struggles that I have is that the shape of the cells is a bit odd so separation based on shape does not work well. and because the intensity varies so much, it does not separate based on that criteria either.
So what do I do to change that?

The picture that I uploaded was not the original, you can find the file below

Kindly,
Lieke

E11219A_gl4-02_s01_AF488.zip (1.7 MB)

Hi @Lieke01,

Still wondering if the image you had shared is a raw image or if it is a projected image.
I tried your image. I have used rescale instensity & Identify Primaryobject module. Please find the screenshot & the pipeline,


Definitely it could be improved with other filters like smoothening & playing around with parameters.
densecell_seg.cpproj (645.2 KB)

Regards,
Lakshmi
Fujifilm Wako Automation (Consultant)
www.wakoautomation.com
For CellProfiler training or optimised pipeline write to,
lakshmi.balasubramanian.contractor@fujifilm.com

Read more on our site.
Yokogawa CV8000 - The Ultimate in Confocal HCS
https://www.wakoautomation.com/products/yokogawa-high-content-imaging

Thank you for your help, but I do have a question. How come that it looks better in the outline image compared to the identify primary objects image? because the outlined one looks very accurate, but the colored one doesn’ t seem to separate the cells that good?
There are still many cells not separated

So, I am still very much stuck on this problem. Is it even possible with cell profiler? or should I switch to another program? could anyone help me out? Here is another example image, I am still no where near getting the cell separation right.E11219A_gl16-01_s03_AF488.zip (8.5 MB)

Hi @Lieke01

Without a nuclear stain, this image will be difficult count. Cellprofiler is amazing at many things but with this quality image you will actually require machine learning to identify the masks. With these masks you can then use cellprofiler to do all the measurements. You could start with random forest or svm pixel classifiers but honestly I would look more into the deep learning models, stardist or cellpose. I quickly ran your image through cellpose to show you how good an annotation you would get.

Best
Lee

2 Likes

Hi @lee.marshall

Thank you for your suggestion. I tried to install cellpose, which was quite a challenge already. When I try to run segmentation the program just gets stuck and wont do anything. Can you maybe specify which variables you started with? Maybe I am doing something wrong there.

Kindly,
Lieke

Hi @Lieke01

Can you please elaborate on what you mean stuck? Are you using the gui version? Do you have a gpu? Images take time to process so you have to be patient when using a gui with only cpu? Is python giving you errors?

Dear @Lieke01,
The outline image is nothing but showing the segmentation & each of the segmented object is just filled with colors. They both are same. You might have to play around with the parameters a bit. You could try playing with the declumping parameters which I have tried with your another image set (PFA pipeline).
If this is not fine, then I may suggest you to try with some Machine Learning approach. For example, you can try in Illastik & use the trainer in Cellprofiler to run for your complete dataset.
densecell_seg.cpproj (645.2 KB)

Regards,
Lakshmi
Fujifilm Wako Automation (Consultant)
www.wakoautomation.com
For CellProfiler training or optimised pipeline write to,
lakshmi.balasubramanian.contractor@fujifilm.com

Read more on our site.
Yokogawa CV8000 - The Ultimate in Confocal HCS
https://www.wakoautomation.com/products/yokogawa-high-content-imaging

Hi @lee.marshall

I figured out that my computer is just too slow to handle this very well. Due to the corona lockdown I am working from home on my own laptop.
I got cellpose to work and I tried tweeking it a bit. This is what I have so far


Sometimes one cell is segmented in to multiple by the program, how do I change that. And in terms of automation, what are my options with cellpose? Do I have to write some kind of script in python?
And how can I do this manually and train the program myself?

Kindly,
Lieke

Hi @Lieke01

You can play around with the object size, but as the model was not exactly modeled on this specific type of cell, I am still impressed with the quality. You can actually train your own cells, I would recommend not training from scratch as the cellpose model is very good so far. They have a great GitHub will all the info you need. Your only difficulty would be created the training set on you samples, but there are many recourses.

Alternatively you can import the masks directly into cell profiler and create filters for object size or intensity to clean up the masks.

Best
Lee