Measuring large SpatialData objects

[pakiessling]

Cool project - congrats on the preprint.

I have a Spatialdata objects with large microscopy images. And polygons / a label image that indicates nuclei which are of interest to me.

How do you suggest getting morphological information for all of those nuclei?

I though about having a generator that subsets the Xarray to the bounding boxes of the segmented nuclei and passes this and a mask to cp_measure.

Then one can have multiple workers which go through the queue.

Does that make sense, do you have experience with such a setting?

[afermg]

I usually do something similar to your xarray generator and then run multiple workers using a queue. That seems to be the simplest solution, as long as your labels do not overlap (which is not yet supported).

@timtreis may have additional suggestions since you are using SpatialData objects.

[timtreis]

@pakiessling and I already briefly talked on the scverse Zulip, but for more public visibility here as well: scverse/squidpy#982 basically implements this exact functionality :)

as long as your labels do not overlap (which is not yet supported).

I solved that in the PR using the associated segmentation mask by dynamically cropping around a specific given cell and then blacking out everything but the cell. So every cell gets an isolated tile, doesn't matter if they overlap.