Masking images

[qu4drupole]

My objective: generate IF images cropped to an area of interest and masked by cell boundaries. Then use images to measure things, like colocalization.

I'm working with MERSCOPE data that looks like this:

Up to this point, I had been using other software to process and analyze the data. I can identify cells in my area of interest, through a merge with "table". From there, I'm not sure what the best option is to generate masked images. I can think of 2 options:

1. SpatialData.query.polygon(). But this is taking forever
2. Pull the Shapely polygons from shapes and create a numpy.array mask which can be used on the raw images. This is fast, but the coordinates (maybe microns?) don't match the image dimensions. SpatialData knows how to convert them, but where is that information stored? Is there a convenient way to extract the polygons in pixel units?
3. ...some other more efficient method to perform this task?

Thanks!

[qu4drupole]

If anyone is curious, here are some options that work:

The best (and obvious) answer to use spatialdata.rasterize() on the selected shapes. Convert the raster to binary.

To do things extra manually, there's spatialdata.transformations.get_transformation(). For an array of polygon coordinates: shapely.get_coordinates(POLYGON.exterior), add a third column of 1 (for the translation). Then the transformed coordinates should be [tranformation matrix] x np.transpose([coordinates]).

[LucaMarconato]

Thanks @qu4drupole for the discussion and for sharing your solution, which is also the one that I recommend.

I add an extra detail regarding polygon_query(). We found that having large collections of points may dramatically impact the performance of the queries. To see if this is the case, we recommend to use spatialdata.subset() to create an in-memory shallow copy (instant copy) of the SpatialData object without the points and then perform the query. There will be a warning about this in the next release: #777.