Curvature (not) accounted for?


Hi Julien/SCT team,

I have a few subjects where they must either have a torticollis or scoliosis (or were really uncomfortable lying in the scanner) - as the end result of the registration is as below:


The coronal view shows the PAM50 and the overlay is the subject’s T1 after registration with the following command:

		sct_register_to_template -i sanlm_structural.nii.gz -s sanlm_structural_seg.nii.gz \
		-ldisc sanlm_structural_seg_labeled_discs.nii.gz -c t1 -ref template -param \

Should I ignore the apparent curvature in the spinal column after registration, as it seems to suggest that something hasn’t worked quite right…?

Best wishes,



Hi Jon,

sometimes the spine is “apparently” RL-displaced wrt. the cord at a given coronal plane (e.g. neck rotation), so I would evaluate the quality of the registration by focusing on the cord (coronal and sagittal views).

if you still observe a displacement of the cord wrt. PAM50 cord, then i would suggest to check the segmentation.



HI Julien,

Thanks for your reply. Rather than the spine being displaced R/L it looks like there is neck rotation (as you say) i.e. the long axis of some cervical vertebrae are not parallel to the long axis of the cord, hence the discs are not perpendicular to the cord when viewed on the coronal. Sorry for labouring this point - just want to check we agree on what the “issue” is.

I guess this might not be a problem if the cord can move largely independent of the spine, but it gives the strong impression of being out of alignment with the PAM50 template. The cord itself appears to be the correct shape and centred on the template for all slices, and the segmentation is good. The only suggestion that something might be off, from the cord’s perspective, is that some of the nerve roots appear to enter/exit at different rostro-caudal levels for a given segment.

Is there any case where too much neck rotation can cause sct_straighten_spinalcord to not straighten the spinal cord enough…? Sorry for the daft sounding question…

Cheers, Jon


we have a neuroimage publication right there :wink:


Good idea! Presumably you could get at the amount of curvature detected by sct_straighten_spinalcord then compare it to some measure of curvature based on the vertebral bodies - I guess the “true” amount of curvature would lie somewhere between the two? Sounds like a job for someone with a maths/engineering background :smiley:


We currently don’t have a good way of assessing the orientation of the vertebrae as most tools are focusing on the spinal cord. But there are a few requests for developing segmentation of the vertebrae, so we might implement this feature at some point.

As you pointed out, you could already access the curvature of the cord, as an output from the straightening routine or process segmentation. Coincidentally, we are currently upgrading the QC report to output centerline fitting and angle between cord centerline and axial slice (Pull request #2299). Example of prototype:

Now is a good moment for feature request!