Support adaptive background thresholding for Xenium data via stain_bg_percentile argument

[katwre]

Hi hi,

I wanted to share my experience running SCS on 10x Xenium data - I'm trying to add it to the nf-core spatialxe pipeline ( nf-core/spatialxe#130 ). I found that the default background threshold (stain_bg_threshold=10) does not work well for Xenium data, since almost all pixels have values above 10. This results in no background being detected and the model failing to train properly.

I hope it's fine that I prepared a pull request to adjust for that (#22). Summary of changes:

• Added an argument to the preprocess function called stain_bg_percentile that allows users to specify a percentile-based adaptive background threshold for stain intensity.
• Added a note in the README and example scripts about using stain_bg_percentile for Visium/H&E data.
• Clarified that the default threshold is for low-range fluorescence images, and percentile-based thresholding is recommended for H&E.

Please find attached a notebook with my test code Test_SCS2Visium.ipynb and the results as a screenshot:

Let me know if you'd like any further changes. Thanks for maintaining SCS!

Cheers,
Kasia