Pipeline module request to check confluency of MSCs

Request to help/guide pipeline module development for raw unstained images to check confluency and cell number per square cms. We will be looking for pipeline module of cell profiler of an tiff image (raw unstained image) of 4X or 10X image. It should automatically calculate the % of confluency, cell size, cell shape and total number of cells per square cms(standard cell culture screening assay). Would it possible to provide the image analysis pipeline to detect the raw unstained images of tiff of 4X or 10X taken images to check the confluency, cell size, cell number, cell shape of adherent mesenchymal stromal cells.

Anticipating positive reply


Hi Sanjay,

First, thanks for your interest. Note that I moved this topic to the Help forum, since you posted essentially the same question in another post there (Human mesenchymal stromal cells).

Now to the biology and imaging recommendations:

  1. Bright-field images are much more difficult to analyze than fluorescent or stained. Our experience is that one can spend significant amounts of effort attempting to segment bright-field images, but if at all biologically possible, that time is would be better spent adding any cellular marker/stain. Please refer to this FAQ for more ideas: viewtopic.php?f=14&t=806#p4488.
  2. Some of your images are jpegs (not recommended), another is a TIF (recommended). Please choose TIF (or any other lossless format) to save your images. From our CellProfiler manual (cellprofiler.org/CPmanual/LoadImages.html):
  1. Your jpeg images are color. CellProfiler can easily convert them to grayscale (using ColorToGray module), however since each RGB color channel has different information, you need to decide whether the color information is useful to you. If it is not (and I suspect it is not given the images I see and your goals), it is best to set your microscope software to save as single-channel grayscale images. Otherwise we would have to pick one of the RGB channels randomly without your domain-specific knowledge. Also, CellProfiler will run faster with grayscale inputs, since the image will need 1/3 the memory.
  2. Any per-cell measures almost require a cell stain that marks individual cells. A nuclear marker is usually the best, since they will be likely separated (not touching).
  3. We cannot know for certain the images’ absoute scale unless you provide a fiducial marker in the image. CellProfiler uses pixels as its scale, and typically the biologist will translate these vales post hoc. We can sometimes guess from information in the image header, but this is not always reliable.

Regardless of the above caveats, I have constructed a pipeline to start you off. This pipeline does well at measuring confluency in these images (including the one from the other post). I have added some notes in the comment boxes above some modules’ settings.

  • Note that this pipeline will only work on single-channel images. I started witha pipeline that had a ColorToGray module to split up the channels, but it was slow and for the reasons stated above, I think you are best to split the channels before CellProfiler.
  • It assumes the files have the characters"green" in them, since I extracted the ‘green’ channel from your jpegs, and renamed your TIFF manually.
  • You can add any measurement modules to this that you see fit.
  • This does not segment cells well, for the reasons stated above. But at least it measures confluency!

Please use this as a starting point, but I think you would best modify the biology side of the images to best help your image processing.

PIPE_MSC_green.cp (4.8 KB)

Thank you very much.

Bright field microscpic raw images are difficult to automate to check the conflency. Would cell profiler helps to quantify the differentiation assay, where we stain the cells with oil red for adipogenesis, AgNO3 for silver staining for osteogenesis, Safranine O staining for chondrogenesis.

With Regards

Very likely CellProfiler could quantify that. Give it a try!

Hi David

I have used this pipeline and tried to modify it a little bit . I am using unstained bright field images for my analysis and as you know they are extremely difficult to analyse. I have reached a point where this pipeline can help me find out the confluency but it cannot separate between individual cells and thus, it is not possible to take individual measurements. I have tried to do that in different ways but it still does not do it automatically. I am trying to differentiate Mesenchymal stem cells (MSC) into neurons and I want to see if my MSC can attain neuron like morphology.

Because of this I have made a simple pipeline to manually identify cell and then take their measurements.
Is there any other pipeline where in I can identify cells automatically in bright field without actually staining them.

Looking forward to hearing from you.

Dhanak Gupta

Hi Dhanak,

As discussed, brightfield images are difficult to analyze. This is our FAQ on the subject:

You can search the CP Forum and find some help, like this:

But I will let you search further.

Another approach would be to take a z-stack of brightfield images and use the MakeProjection module with the Variance option.
If you look at the Help for this, you will see that the Variance option implements Selinummi et al’s (2009) approach to BF images. Give it a try!

Good luck,