How to separate IHC images and creating new pseudo IF imagine?

Hi. I have recently started learning Qupath and I am very entusiasted with this program. I read the script from Mr. Pete Bankhead @petebankhead and watched the videos from the course from February this year. I am very interested in processing two IHC-DAB scans into one multi-channel pseudo IF image, just as shown by Mr. Pete Bankheadon @petebankhead day 1 of the course (4:33:00).


I would like to combine two IHC images (CD68KP1 and CD30) from two subsequent cuts and see if such an image will be a reliable object for analyzing the distance between these two types of cells.

Can I count on help in recreating this process.

I will add that I am a resident of pathology and I do not have programming experience at all, but I like to learn. I will be grateful for any help in finding learning materials . I will not be offended for “dummies” style guides - it is my first Topic here :slight_smile: .

Hi @Sokol.kamil, I’m glad you’re enthusiastic about QuPath!

That session of the workshop was on ‘what I’m working on / where things are going’ and I showed a (rather complex) script to illustrate that QuPath can support this kind of transform in principle, and it is likely to become part of the software in the future. However, it’s not something I have had any time to work on since February, and it’s not part of the software yet.

1 Like

that’s a shame :frowning:
I hope you will be able to introduce yourself in the near future.
Thank you for the quick reply.

I hope so too, but realistically it is likely to take some months at least – there are lots of other things to work on first.

However, I strongly suspect that it wouldn’t really help very much in your case since the differences between subsequent cuts are likely to be too big for the simple alignment to cope with. QuPath doesn’t do image registration (a frequent question…), so any transform more complex than ‘shift and rotate’ that is needed to align sections would need to be done elsewhere anyway.

1 Like

I’ve worked with a few of these in the past now, and I would tend to agree, depending on what your markers are. If you have a general tumor marker (or other general marker for… something) in one section (vimentin, cytokeratin) to determine large areas, and a specific cell marker in a second section, you can get a fairly good idea of “are my cells closer to or farther from my tumor” and similar questions.

You cannot really perform coexpression studies in most cells, 5 micron slices are just too big and while many of the cells are the same in any two slices, many also only exist in one slice. The lack of overlap between many cells will do weird things to your data.

Two independent cell types would be somewhat borderline, if you knew that the two markers would never overlap. With two slices you get an increased chance of fake “double positive” cells (or paris of cells with a “distance” of nearly 0 microns).

For studies looking at cell positions vs each other, there are options like true immunofluorescence, or the strip-stain methods like MICSSS: https://www.mountsinai.org/about/newsroom/2016/mount-sinai-researchers-develop-simple-method-to-characterize-immune-cells-in-tumors
I assume there are other methods similar to this as well. In the case of the strip-stain, at least all of the cells are the same cells between the two stains, so extracting the DAB “channels” and overlaying them would make sense.