Allow per-channel fixed-value background subtraction in illumination correction

[loredanacantoni]

Currently fixed-value background subtraction uses the same background value across all channels. However, background-values vary between channels, which can lead to inaccurate corrections.

Possible implementation: allow fixed-value background subtraction for each channel individually.

[jluethi]

That's a good feature request. It could go together with the option for users to actually provide darkfield images to be subtracted, not just flat values per channel.

Is this a major blocker on your side @loredanacantoni at the moment? If not, I suggest we take this into consideration for the ngio refactor of the illumination correction task (cc @lorenzocerrone )

---

A question about

background-values vary between channels

Does your average camera gain vary by channel? Or would you want to subtract more because of e.g. background staining? We've mostly used this to subtract the background gain value of the camera for a given channel and that has typically been constant across channels (thus, varying background value by channel hasn't been a feature request so far)

[loredanacantoni]

We noticed, that on antibody channels we often have a much higher background than on nuclear stain channels. I believe it is mostly due to background staining, rather than differences in camera gain. However, the image quality could be improved if we could reduce the background.

Darkfield subtraction would indeed be a very elegant solution. For us it's not a major blocker right now, since normalization with the correction matrix removes the vignette and already improves quality considerably.

[jluethi]

Sounds good!

For more complex background patterns like this, the BaSiCs approach could also be meaningful: https://github.com/peng-lab/BaSiCPy

It's available as a Fractal task here: https://github.com/Apricot-Therapeutics/APx_fractal_task_collection