Hello, I am currently conducting a lineage tracing experiment where I need to count GFP+ cells in multiple compartments of ductal epithelium (frozen tissue sections of an inducible Cre mouse with mT/mG reporter).
Attached is an example image, I need # of GFP+ cells in each of the circled compartments counted separately and also a count of GFP+ cells along the epithelium (straight yellow line indicates this area for an example). (Yes I am aware the number of GFP+ cells in this image is low, this differs wildly between control and experimental groups)
Up to this time I have been doing these counts manually, but with the number of sections I need to count this is becoming unfeasible.
Could someone point me in the direction of Image J tutorials so that I can learn how to automate this process? (I also have access to Halo software)
Thanks for your time!
Kyle
If you want to automate the method more on imageJ, the first thing I would suggest is get a higher resolution image from a let's say 40x objective. If you want to select the GFP positive cells only, select the green channel (using split channel option) and use that image for your segmentation.The next steps are finding an appropriate segmentation method followed by object detection (each segmented region is counted as an object), and then you can do a variety of actions like object count, object area, perimeter, roundness etc. You need to optimize your segmentation a bit for the first couple of images.
Hi there,
I agree with Biswajoy Ghosh ,
First thing you should try is to get better resolution pictures. Otherwise is going to be difficult to automatize the process...
Another thing that could also be helpful is to use (if spectrally possible) any kind of nuclear counterstaining to be sure that the particles are "n" cells.
Cheers,
J
I do have higher resolution images, wanted to just give an example of what I am working with. I am using DAPI as well, however one thing making things difficult is the fact that the nuclei are either close together or stacked on top of one another, which I am sure will cause problems with the software identifying individual cells.
Biswajoy, could you possibly point me in the direction of a good tutorial for the segmentation, object detection process using Image J?
Thanks for the rapid replies!
Kyle
Hi Kyle McAndrews After looking at the image I have a few comments:
a) the tissues are too dense to be used for automation as it will lead to more errors.
Solution: As I mentioned earlier, you will need a high NA objective image that captures (I) a smaller spatial field and (ii) a thinner section of the tissue in confocal mode. This will remove overcrowding and render the image usable for segmentation with good accuracy.
I learned that 60- 70% of your problems will be solved with the right image.
For splitting color channels (ImageJ>load image> Image/Color/Split Channels )
As for good tutorials, there are several on segmentation /cell count : https://www.youtube.com/watch?v=82N-eIPqnwM
https://www.youtube.com/watch?v=D1qBaFwuF4E
https://www.youtube.com/watch?v=jLd2I2adQtw
https://www.youtube.com/watch?v=zwjXbeiviH0
I am sharing several videos as you will need a combination of one or more ways to suit your specific application.
and refining segmentation (morphological operations)
https://www.youtube.com/watch?time_continue=285&v=gF4nhq7I2Eo&feature=emb_logo
documentation: https://tinyurl.com/y8j82hpp
As J. Ramirez-Franco mentions, you will also need the nuclear stain (which you did) for total cells. Overlapping of the nucleus (due to a thicker section of tissue being imaged) can be reduced by the solution I mentioned. If the problem still continues you can use the 'watershed segmentation' of binary images in ImageJ. you can read the basics from here: shorturl.at/uJMX3. There are some more advanced packages available for more interactive experiences like https://imagej.net/Marker-controlled_Watershed, and https://imagej.net/Interactive_Watershed
Thank you Biswajoy. I will try some thinner sections on my next tissue blocks. I have not cut thinner than 8 uM before so hopefully I am still able to section that that thin without problems.
I appreciate your response!
Kyle
Hi again,
I don't think the problem is the thickness of the sections, but the thickness of the optical slices. Try to get, as Biswajoy Ghosh said, smaller fields of view with an objective with larger NA (ideally I think it would be 63x) and thiner optical slices in confocal mode.
Lets say one of the circled compartments of the first image per field of view. Use the appropriate pinhole (1Airy Unit) and the optical sections shouldn't be thicker than (around) 0.5 um (probably less).
At this magnification, nuclei counting with watershed segmentantion (as mentioned above) works super well (for stacked/piled nuclei).
Cheers,
J
- Badges
- Science topic
- Similar topics
- Biological Science
- Histology