Analysis Hierarchy

Hey folks,

I am trying to geht QuPath running for my purposes and think I could get some conceptual help from you guys:
I am working with (multiple) TMAs and AEC as the IHC chromogen. Overall goal is to calculate number of positive cells for any given stain in an area of interest (tumor vs stroma) and batch analyse all TMAs within the same project.
So far I have manage to

  1. set stain vectors, pos. cell detection parameters
  2. de-array TMAs

I am joyfully reading the posts from Research_Associate like:


as well as Pete´s blog posts and the Youtube videos.

However, I am struggeling a bit currently to define a logical hierarchy of actions for my workflow like “what to do first”:

Idea currently is:

  1. Creating project
  2. Importing all images or keeping TMAs individually ?
  3. Setting stain vectors equally for all images
  4. De-Array TMAs
  5. I guess I don´t need Tissue Detection since I am working with TMAs.
  6. Area identification / Classification: As I´d like to get pos cells in a given tissue category (tumor / stroma), I imagine this needs to come first. So would you use Reasearch_Associates workflow with the SLICs or the pixel classifier to get tissue areas? Would you use Analyze->Calculate features for this ?
  7. Positive cell detection
  8. Analzye all TMA cores / all TMAs
  9. Use sth like https://petebankhead.github.io/qupath/scripting/2018/03/05/script-annotation-results-merge.html
    for merging annotations and exporting

Thanks for your input

TMAs are, well, notorious might be the wrong word, but generally very difficult to batch analyze. Generally you are either looking at staining for different proteins, or different types of samples/tissues, either of which makes running a script on a whole project very difficult. For TMAs with multiple tissue types in the same TMA, even finding a single script that works for a whole TMA can be almost impossible without treating each sub-group of TMA cores differently.

That said…
I’m not sure what you mean by 2. If you want to try to run a script on multiple images, you may as well put them all in the same project. If you have Stain/Negative control pairs of TMAs, then it might make sense to keep one project per. If nothing else, it would make interpreting the exported results much easier.

  1. Tissue detection automatically runs within all available cores when run with a TMA present. It can be very useful if you want cell density as opposed to positive percentage, as the whole TMA ellipse area is often not filled with tissue, which would lead to underestimating density. On the other hand, negative lung tissue often fails completely with settings that work for most other tissue. YMMV.

  2. For the moment, SLICs are probably it, though again you will run into problems if you try to generate a classifier across multiple tissue types or stains. Often times a classifier trained to find a tumor marker will just find all dark areas… which may not always be accurate. The pixel classifier will be better for this, I think, once it works across projects, but for now it doesn’t.

You will pretty much always need a manual step after 4 to adjust the TMAs before running anything else. The dearrayer isn’t perfect, and once you run anything, the TMA is locked into place and cannot be moved (as far as I know) until you delete and recreate it.

Might be overly pessimistic here, as I worked with some weird TMAs, hopefully yours are more normal, but would need images to say more. Overall, that sounds about right.

Thanks Research_Associate,
@ 2. : In general I have 6 TMAs of NSCLC tissues (“only” different histological subtypes [Adenocarcinoma, Squamous Cell Carcinoma]) and 1 TMA of multiple malignancies. All stained with 5 different antibodies. So far I guess I will analyze the samples independently by stain and hoped to analyze the 6 NSCLC TMAs together from within the same QuPath project file. Therefore creating QuPath project files per stain. The TMA with multiple malignancies is probably going to be analyzed apart from the rest.

@[quote=“Research_Associate, post:2, topic:29602”]
You will pretty much always need a manual step after 4 to adjust the TMAs before running anything else.
[/quote] : I have already manually optimized the grid and would like to batch analysis to start from this point.
@5: For the beginning, I´ll be happy with positive percentages.

Also,
would it be possible to use Classify -> pixel classifier for region identification/classification, turning those into objects and then applying positive cell detection in those?

Technically yes, but you would need to create a new one for each image.

OK. Thanks guys. That helped for my decision.

1 Like