Possibility of using target tissue as information for network configuration

[vcasellesb]

Hi,

I was thinking about how to improve my segmentation algorithm (it uses nnU-Net), which segments nerves on CBCTs (images of the mouth), and I thought about ways to improve discriminability between the background and nerve (they have very similar intensities). Since I also want might want to use nnU-Net to segment teeth, I was thinking about wether it would be possible to adapt the normalization/preprocessing scheme to maximize discriminability between tissues that usually have high/low intensities.

I am in no way an expert on the subject, so what I'm saying might be impossible, not make sense or it is already done in a similar way in the current code, but I would like to know if that's the case, and if it's not, I'll try to look into it.

Thank you very much.
Best regards,
Vicent