How to handle fluorescence images of very dense tissue?


when analysing quite dense tissues as spleen or tonsil we find it hard to set parameters for cell detection. We have been playing around with nucleus parameters and ended up setting background radius up to 8 , keeping median filter radius at 0 and setting Sigma to 1.3. Results differ a lot, depending on these parameter values, of course. What could we consider to improve our results, when cell groups are just no be separated into single cells? I attached an example which only 3 groups of cells(more included) .


On the non-QuPath side:

  1. Hoechst washes instead of DAPI in the mounting media for lower background/easier separation (thanks LJI and @smcardle !)
  2. Thinner confocal slices, confocal slices at all, or thinner tissue slices, as I can see enough variation in staining to suggest some cells are “below” your plane of focus. *with confocal, potentially higher pixel count as well.
  3. Within QuPath, lowering the sigma is about all you have access to. QuPath also does have access to ImageJ, so if you can segment these in ImageJ using a larger number and combination of filters, that might be the way to go.
  4. CellProfiler, Visiopharm, etc frequently have far more options for cell segmentation, since they were designed for a greater variety of sample types, rather large, whole tissue.

If you have only single fields of view like this, you may want to go the ImageJ route first. In addition to the Macro Runner dialog, you can also insert a working macro into a script:


Oh, and one final option that was covered in the guide, here, is to convert the annotation object into a cell and use subcellular detection, which is based off of peak intensities, I think. So you have some separation options there. Might be worth a shot.

That looks like a section I need to fill out with more specifics and links, though, so I’ll try and get to that later today.

1 Like