Limitations of Qupath

Can anyone tell me what the primary limitations are of QuPath if you thinking of using it to quantify protein expression levels on H-DAB stained tissues and using it to quantify cell count of fibroblasts in skin samples. E.g. sensitivity issues, general inaccuracies?

Thank you :slight_smile:

Other people may disagree with me here, but I would say the major limit on quantifying H-DAB stained tissues, in QuPath or any other analysis software, is the data itself. Sensitivity and specificity of your antibody are determined by your sample preparation protocol; noise, resolution, and evenness of illumination are all determined by your microscope and camera; H-DAB intensity is known to be non-linear with protein concentration in almost all circumstances. The choice of image analysis software is more about speed and ease of use than “accuracy”- nearly all available software will give you similar answers to easy questions.

For your described application, the only difference I would expect to see between different programs is how the fibroblasts are segmented. QuPath uses nuclear staining + expansion to determine the boundaries of cells. This is computationally fast and works well for many tissues, but is often inaccurate for elongated cells like fibroblasts. Other software packages take different approaches to cell segmentation, which could be better or worse, depending.


Adding to the great post by @smcardle, I would add that the problem might lie in what you call “Quantifying Protein Expressions”.
DAB Stainings and general light microscopy is about “Where your staining is” not so much about “How much staining there is”. I second and emphasize on the fact that DAB has many limitations in terms of correlating it to protein concentration.

The “Best in class” you could achieve is classifying your cells as low, mid and high expressing, and base your analysis from there, with quantitative concentrations coming from indirect methods like Western Blots and the likes. The issue there is that based on the protocol and tissue (and time of day, and horoscope… :slight_smile: ) your thresholds for the different classes might vary from image to image and that has to be mitigated…

QuPath has pretty much no limitations to what it can or cannot do, provided the data is clean and you’re willing to get your hands dirty with some light scripting to get exactly what you want in a reproducible, as automated as possible, way.

Feel free to share images or further questions!



Hi Oli and Sara,

Thank you very much for your insights. I am quantifying expression of PLOD2 and NICD in the dermal and epidermal compartments of skin sections, could you think of any particular limitations regarding the use of QuPath to quantify the expression?



Hi Dhylan,
I think we need to know more about what type of limitations your are thinking about. Are you looking for limitations of QuPath analysis as compared to doing the image processing in similar software (ImageJ, CellProfiler, etc), or as compared to a pathologist qualitatively evaluating your samples or as compared to some sort of platonic objective truth?

1 Like

Hi Sara,

Apologies for not being more specific. I’m looking for any limitations/drawbacks more from a pathology perspective i.e. qualitatively and quantitatively evaluating the samples - any drawbacks you can think of in QuPath regarding protein quantification in this sense. Also, any objective opinions would be helpful too!

As an additional thought, I am also thinking of investigating the binding of NICD (Notch Intracellular domain) and its associated factors that form a transcription initiation complex that regulates cell migration. I am thinking that a DNA binding assay would be useful to assess its activity, but am unsure as to what the best approach is, perhaps you have insights into this as well?



I suppose the worst drawbacks are probably the non-linearity already mentioned, and the fact that it is… brown. One of the worst colors to work with anything else, including blue for hematoxylin. At least if you use a red or green dye, you are not using the same wavelengths of light or the same detectors on the RGB camera. Brown blots out pretty much everything once it is too dark, and even when it is faint, it interferes.


I think the question is very hard to answer because so many things are interrelated, including

  • biology
  • experimental design
  • tissue preparation / thickness / size / sampling
  • staining
  • imaging technology / image type (including fluorescence, brightfield, something else…?)
  • image quality / compression / resolution
  • method of analysis, e.g. using software or not
  • output metrics (e.g. a cell count, H-score, Allred score, density, some intensity-based value)
  • how the output metrics are interpreted / overinterpreted

Then, if QuPath or any other image analysis software is used, there remains no one method for ‘protein quantification’. There will inevitably be variation and limitations in

  • the algorithms and functionality within the chosen software
  • the way in which the user applies the software

Unpicking all of these would be hard work for anyone, even if they had an in-depth knowledge of what kind of data you are thinking of quantifying – and really impossible without this information, or even example images.

If you would like anything more specific than the excellent answers already given by others above, I think the question needs to be more precise and specific. My initial feeling as to whether there are any limitations/drawbacks when using QuPath is ‘yes, probably hundreds!’, but the overwhelming majority aren’t limited to QuPath and not all would apply in all cases.

For that reason, I think it makes most sense to consider limitations in comparison to some other method of assessment.

With that in mind, can you explain for context what is the purpose of the question?

And could you list the relevant limitations/drawbacks of any other method or evaluating the samples (e.g. visual assessment)?

Since it’s free and open source, with lots of docs and tutorials online, I’d recommend just downloading it and making your own judgement :slight_smile:

You can also see how it has been used by others in the list of publications.