Hi,
I have noticed that the array in the saved segmentation file (.seg.nrrd) is not really one-layer-per-label. Instead, non-overlapping labels are merged into one layer, while overlapping layers are kept separately.
I find this quite clever, but for convenient echange with deep learning (specifically MONAI Label), I would like to store it as a one-layer-per-label representation with exactly the same voxel grid size as the source image volume. The latter seems to be the default behaviour in the latest Slicer versions, but I couldn’t find a discussion on “one-layer-per-label” (perhaps I don’t know the right search term).
Btw, I know that MONAI Label allows overlapping labels in principle… but I am playing around with a 2D radiology app right now, and I need to manually change the storage behaviour of .seg.nrrd files. Would be great if there was a global flag (e.g. in Application Settings) that could switch that on.
I also contacted the authors of the MONAILabelReviewer module, who contributed overlapping-labels support in MONAI Label, but they also didn’t know how to avoid the layer-collapsing behaviour in .seg.nrrd.
Thanks for your advice in advance!
Maybe you can just standardize on the newer version? There will be a new 5.4 release before too long that will be essentially the current preview version with a few tweaks.