Positive cell quantification for membrane and cytoplasmic markers

[Carlos Moro]

Hello,

So far I hadn't tried QuPath quantification (% positive cells) with membrane and cytoplasmic immunohistochemical markers.

One introductory question, when "Estimate stain vectors"  a pop up says: "Modal RGB values 252, 253, 251  do not match current background values - do you want to use the modal values?". What does it mean practically? How to decide between Yes / No?

After tuning a bit cell detection (as in screenshots and previous threads), checking "Exclude DAB (membrane staining)"  and  "Score compartment:Cell DAB  OD mean" following results come.

Slides 1-2: Membrane marker

Slides 3-5: Cytoplasmic and membrane marker

Slides 6-8: Cytoplasmic and nuclear marker

What do you think? Would you suggest some alternative method/param setting for this task?

Also I wonder, what's the exact difference between "Score compartment: Cytoplasm" vs "Cell: DAB OD mean? I didn't see much difference with these images. 

Best wishes,

Carlos

[micros...@gmail.com]

The score compartment is actually very much what it says, you can see the values that it is looking at for thresholds when you select any given cell (hierarchy tab).  Cytoplasm is the area in the cell expansion, and whole cell includes the nucleus.

How useful any individual measurement will be will be dependent on how large your nuclei are, how much cell expansion you allowed during cell generation, how dark your staining is, etc.  A very large or dark cytoplasm won't see much difference vs "cell mean" because there won't be as much nuclear contribution to the mean.  And if your negative cells are very negative, almost any low enough threshold will pick up the same set of positive cells either way!

Provided you have checked "Make Measurements" you can see roughly where you will want your cutoffs to be by using Measure->Show Measurement Maps.  Drag the slider around until all of the cells you think should be positive show up as red when viewing the metric of interest, and use that value as your cutoff when classifying.  

One of the problems with membrane staining, though, is that at this point I don't think the cell expansion is "intelligent" and so you can very easily pick up neighboring positive membrane staining.

[Pete]

Color deconvolution involves normalization to background values; the popup in the estimate stains command is offering to set background values based on the most frequently occurring values within the selected region.  Assuming the selected area contains background, and the suggested modal values are reasonably high, this tends to give a good background estimate.

Now an admission: as far as I remember, the cell detection command doesn't actually make use of the background values set here and defaults to values of 255 per channel instead.  Also, your background values are very close to 255 anyway.  Therefore for both these reasons it is unlikely to matter very much at al whether you accept the modal values or not.  I'd tend towards keeping the defaults in your case, so choosing 'No'.

I agree with microscopyra for all the rest.  I would just add that there is a command in the same menu as Cell detection called Cell + membrane detection that does try to make use of membrane staining to constrain the cell expansion more intelligently.  It has the benefit of making measurements of membrane staining separately, and is capable of identifying cells based on membrane staining alone even if the nucleus is not evident - while still given fairly sensible cell estimates when only the nucleus is visible and no membrane staining is present.  On the other hand, it is less flexible than the standard Cell detection (I think it is only for DAB + hematoxylin staining).

Cell + membrane detection is probably worth a try, since it can sometimes lead to a better result.  But it is trying to do a really difficult job, i.e. detecting cells without necessarily being able to assume that membrane staining is always present, nor being able to assume that the nucleus is always visible.  Therefore results can be mixed depending on the image and staining; sometimes better, sometimes not.  Personally, I have only used it myself on rare occasions.

[micros...@gmail.com]

Interesting, I had not actually tried that particular tool.  I have no idea what I thought it did, but I will have to give it a shot!  And even if it does expect Hematoxylin+DAB, assuming it gets those vectors from the image settings, you can make "DAB" whatever you want it to be!  In this case, maybe your AP vector.

Side note: I have noticed some interesting dependencies between some of the tools and the actual label on the vectors, which I haven't entirely mapped out.  For example subcellular detections gives you 2 channel options in Image Type - H-DAB as long as the first vector isn't actually called "Hematoxylin."

[Carlos Moro]

Thank you so much, I'll try the good advices and "Cell + membrane detection"!

Best /

Carlos

[Carlos Moro]

Hello,

Thank you again for the good advices, it was very interesting to see how the different measurements (cell, cytoplasma, and membrane DAB OD) differ and using the measurements map to adjust the threshold!

Attached some pics of the different modalities. In this image, where membrane staining was very partial, cell detection was comparable by using (slides 1-2) "Positive cell detection" and (slides 3-5) "Cell + membrane detection". But the last one with checked "Make measurements" made Membrane DAB OD measurements appear in the Measurement maps. Comparing them with same DAB threshold  % positive decreased from membrane > cytoplasma > cell, which makes good sense. 

Probably using the more restrictive "Cell DAB OD" alleviates eventual multiple limiting-cell detections due to cell expansion...

Best wishes,

Carlos

[micros...@gmail.com]

Nice work, and a nice sample for this as well.

I wouldn't call the full cell measurement "restrictive," so much as "lower" due to the lack of DAB in most of the nuclei.  In this case you would (probably) get a similar %positive change by increasing the threshold between Cell and Cytoplasm OD measurement tests by something approximating the percentage area of the nucleus. So maybe 0.12 to 0.15 for the cytoplasm DAB OD.  Or if you wanted to get really creative, a separate threshold for every cell based on the nucleus/cytoplasm ratio should fit even better.

I am not actually certain what the difference between the membrane and cytoplasm DAB OD measurements actually is.  I don't  have any samples quite like yours, so I wasn't able to determine if it might be some sort of thresholded area value (only measuring parts of the cell over a certain DAB OD) or maybe pixels within X distance of the edge of the cell.