Canal or not canal ? How to segment on axial acquisitions

[abelsalm]

I started slecting the data I really wanted to train the model on, for instance to remove images with motion artefacts.
But I realized there is still one dataset I'm really struggling with : dcm-zurich.
On those axial acquisitions, I can't really say if some parts are associated with the canal, or not.

For instance on this subject, here is the image and the actual segmentation (not really good I think)
sub-878584_acq-axial_T2w_050_0000.nii.gz
sub-878584_acq-axial_T2w_050.nii.gz

Especially on the disks levels I don't know how to segment, for instance here :

Taking or not taking the green part ?

Or here : I don't jnow if the yellow circles better than the green, or neither are really accurate ?

May be you would have ideas @SomeoneInParticular @sandrinebedard @maxradx ?
Those are not particular cases, the whole dataset is interesting for dcm cases and almost every image creates the type of doubts....