the registration algo is unreliable (because partial seg in source vs. complete seg in the destination)
the output softseg will be ulgy
Solution?
We need to crop 2-3 slices from the registered coverage image (eg: sub-brnoUhb05_acq-T1w_MTS_ones_reg.nii.gz), and then add in the analysis pipeline a multiplication of the registered segmentation and of the registered coverage image.
How do we detect the slices to remove? we could create an algo that finds the minz/maxz that have at least one non-null voxel in the registered segmentation, and that will be the minz/maxz from which to crop 2-3 slices.
sandrinebedard
commented
1 year ago
In relation to https://github.com/sct-pipeline/contrast-agnostic-softseg-spinalcord/pull/32, the new results produce partial images in the destination (T2w) space (see last slice):
This is problematic because:
Solution?
sub-brnoUhb05_acq-T1w_MTS_ones_reg.nii.gz), and then add in the analysis pipeline a multiplication of the registered segmentation and of the registered coverage image.