Subcellular detections on several cell types

Hi, I have samples that are stained with a cytoplasmic neuronal marker (NeuN), a cytoplasmic glial marker (S100b) and which contain subcellular signals in several other channels (RNAscope staining). Here’s an example with NeuN + subcellular signals:

Screen Shot 2020-04-27 at 08.42.21

I’d like to segment the neurons and glia based on their cytoplasmic staining and then run subcellular detections on cell body (not nuclear) outlines. All of this would be done after running automated tissue detection. So the hierarchy would be: Tissue annotation > two different parent object classes (neuron and glia outlines) > several child classes (RNAscope signals) for each of those.

I am wondering what the best way of doing this would be in QuPath? Multiplexed classification (as described here: probably won’t work, since it relies on nucleus/cell detection first, before assigning them to classes. I guess a possible option would be to run neuron vs glia cell detection + corresponding subcellular detections in two separate steps, export measurements separately and then combine the data afterwards…. But I’m sure there’s a better way that I’m missing as a QuPath novice (e.g. detect neurons first, preserve annotations, Preserving rounds of cell detection?, detect glia, then run subcell. detections).

Many thanks!

If you only want cytoplasmic stained spots within your neuronal markers, you are probably best off with multiple rounds of cell detection and exporting the results. If you want to keep both together for visualization, then as you linked, you could save one and then reload it after running the second detection.

Other options could include using Simple thresholding or the Pixel classifier to create objects based on your cytoplasmic markers, which would each have the trained classification.

In both cases, you would need to make sure the object was considered a “cell” object prior to running subcellular detection. As far as I know, that can still only be run in CellObjects. In the case of the Cell Detection command, if you have 0 Cell expansion, the resulting objects are not considered cells as they don’t have a cytoplasm. Similarly, you would need to build a cytoplasm onto any pixel classifier objects, even if it is only one pixel thick, before running subcellular detection on them.

The pixel classifier might end up being more accurate in terms of cell borders, but would probably be more difficult to code.

Thanks for the feedback, will explore those options!