Cell detection and histomorphology parameters, revelation with histogreen


I use your software recently, it looks very well-made, with so many opportunities.

And for a research work I would like use it, to analyse the morphology and immunohistochemistry profiles of renal tumors on tissue micro-array slides. I would like to know what are the best parameters to correctly analyse these tumors, because they are very heterogeneous (in shape and in appearance). I tried to modify the nucleus and cell parameters, but the software is struggling to properly identify and cut out cells. How can I modify the parameters of the “cell detection” ?

For this I use whole slide in qptiff format and pictures in jpeg format.

And, another question, I made immunohistochemistry with histogreen system of detection and not DAB, will the software correctly detect the marking? If yes, which settings should I choose ?

Thank you in advance for your answers !

Once the forum lets you post pictures, those would be the most helpful in determining a method for analysis. Unfortunately, there are too many different ways to analyze things, and too many different tissue types and stains to really go into many details unless I was familiar with more of the specifics of the study.

One thing I can tell you is that if you have a single stain that is not DAB, you can simply set the color vector for DAB to the new stain. How effective that will be will depend somewhat on your stains, though. A blue stain and blue counterstain would not be magically resolvable :slight_smile:
Also in video format: https://www.youtube.com/watch?v=IpCDnPnFvDc&list=PL4ta8RxZklWk_O_Z7K0bZlhmHtaH73vlh&index=2

Sometimes the staining just isn’t clean enough, or there are too many wildly different shapes (dense elongated nuclei and extremely large nuclei with only intermittent nuclear staining) for the default cell detection to work on all cell types. You may need to script something a little bit more complicated if you are in one of those edge cases.

JPEG in general should not be used for analysis on its own (there will be no metadata, built in blurring), though it or JPEGXR are used to compress tiles of whole slide images in many cases. I only include this because I am not too familiar with the qptiff format and was not sure if your whole slide images are JPEG compressed, or you are trying to do analyses on stand alone JPEGs (if so, compare your Image tab settings to see why this is a problem!).

Hi Mike,

Thanks you for your answer ++ And sorry for my late reply…
I uploaded three pictures, 2 of HES, in JPG format, for the morphology analysis that I would make, and the third, an immunohistochemistry slide with Histogreen revelation, that reveals the presence of anti-COX VI antiboby.
Like you can see, the two HES slides present so many differences, and I would like to see if QuPath software could identify them more correctly. But I have some difficulties to find good parameters…
For the format, I take these pictures with the Leica camera of my microscope, but I can save them in different formats like Bitmap, PNG, TIFF, JPG and JPEG2000. Do you think one of them is better for analysis ?
For the IHC I I have not tried yet, I will try it as soon as possible.

Generally TIFF will be best, and hopefully your software can save the images with accurate metadata. For TIFFs, even if your software doesn’t save the images with metadata, you can go into FIJI and edit the pixel data directly. Almost any analysis in QuPath is easier with pixel sizes, but you haven’t been too specific about what exactly you want to do.

You did mention cell detection, and two things you will want to do are:

  1. Make sure you have pixel size metadata
  2. Set your stain vectors correctly for a given image.
    If you haven’t gone through them, the first two videos in Pete’s second link here should help.

In your case I could easily see the second two stains working in terms of finding cells with or without cytoplasmic stain. Unfortunately, currently cell expansion is based on nucleus shape only, and not surrounding stain. So you will not be able to get cell shape information if that is what you are interested in. The first image seems the most problematic since there is not much information close to the nucleus, and expanding the cytoplasm out too far from any given nucleus will overlap into nearby “real” cells.