Importing cell segmentation performed elsewhere

[degrainger]

Hello,

Thanks very much for developing and sharing this package, I think it is a great step forward in analysing Visium HD data.

I was wondering if you could advise us on how to import cell segmentation performed outside of bin2cell? If, for example, I wanted to manually curate the cell segmentation and then import it for segmenting Visium HD 2um bins.

I would obiously need to perform this on the rescaled H&E image, save the file, import it, convert it to .npz format. Could I then just plug it in to b2c.insert_labels()?

If I've understood mpp correctly I would just set it to the pixel size of my input image?

Thanks for your support, any tips would be appreciated. Best wishes, Dave

[ktpolanski]

Yep, you've got the gist of it down. Set the mpp to whatever you set it to when you made the rescaled H&E image, make sure you're pointing at the correct set of spatial coordinates (the H&E image function crops out the part around the actual captured tissue bit and saves it to the object as .obsm["spatial_cropped"] by default as shown in the demo) and you're good to go. As for the .npz contents, it's just a scipy sparse matrix matching the shape of the rescaled H&E image with 0 as the background and integers as the identified objects. You can load one made/saved by bin2cell via scipy.sparse.load_npz() and take a look around if you want to see.

[nadavyayon]

I just to make sure, I think you got it like kp said, but i would emphasize that i would use the rescaled image from b2c to run your manual segmentation to make sure there isn't some weird precision issues of how images are rescaled.

So pipeline is: read object with b2c -> rescale the he to whatever resolution you want -> take the rescaled image and do your thing -> generate the masks -> convert it to .npz -> read back with insert labels.

In other words I would be careful if you plan to take your original image, scale it externally to the same mpp and then feed it back to b2c