Apply segmentation to H&E images

[maiskovich]

Hello, thank you for the great package.

I first tested the tool with some of the provided images, and the segmentation worked very well. I was expecting to get similar results with H&E images. I tried cellsam_pipeline method with two different public H&E images, and I got quite bad results. These are relatively small images (3000x3000px). I tested using use_wsi False and True, when I set it to true, I tested different block_sizes (256, 512, 1024), but still most of my cells are not being recognised.

This is how my image (public H&E) segmentation look like with use_wsi=True and block_size=1024 (actual image is bigger, this is just a crop of some of the results):

Is there another kind of setting that is suggested to use when working with H&E images?

Thank you

[rossbar]

Thanks for raising @maiskovich - that looks to me like a scaling issue. Perhaps you can try the gauge_cell_size option (i.e. gauge_cell_size=True) along with use_wsi=True. Note that you'll need the latest version of cellSAM for this, so make sure to pip install --upgrade git+https://github.com/vanvalenlab/cellSAM@master before trying!