Mistake in the dataset.json

Hi,

In the given dataset.json file, the channels are both named "CT". This triggers in turn the CTNormalization normalization scheme for PET images in nnUNet, which might be bad. I changed it to "PT", to recover the "ZNormalization" normalization scheme for PET images Keep an eye out for it if you hadn't noticed it before

Regards Zhack

Hi Zhack,

That's actually not a mistake. This is set on purpose. If you assume the standardized uptake value (SUV) does what it should and represents standardized physical quantities, a global normalization scheme is a valid assumption. However, one could argue that this assumption does not hold, especially in a multitracer, multicenter setting. So, trying out new approaches might be beneficial. Thanks for bringing that up!

Best Jakob

Dear Jakob, Thanks for the clarification!

This makes sense indeed, and I think both methods do have their pros and cons

My reasoning for preferring ZNorm (no big trade secret ;) is that you may have patient specific variations in SUV values / distribution, linked to their activity between injection and acquisition, glycemia and other physiological specificities.

Best, Zhack