Hi Sergio,
I came up with something that works reasonably well. Here’s the syntax:
sct_register_multimodal -i S1_T2.nii.gz -d S1_CT.nii.gz -ilabel S1_T2_labels.nii.gz -dlabel S1_CT_labels.nii.gz -iseg S1_T2_seg.nii.gz -dseg S1_CT_seg.nii.gz -param step=0,type=label,dof=Tx_Ty_Tz_Rx_Ry_Rz:step=1,type=seg,algo=affine
Before registration:

After registration:

Now, I’ve only tried one combination of parameters (because the CT data is large, it is slow to run), but I encourage you to try tweaking step=1. A few things to try: change the shrink factor (to 2 or 4), increase the number of iterations, add smoothing. I wouldn’t play with non-affine transformations, because the information content between the CT and MRI vertebrae segmentation is quite different, so you would end up with non-robust results. While you do this optimization I encourage you to visualize the output of the step=1 (use the flag -r 0
to keep the intermediate results), as shown below (red: before step1, green: after step1, white: CT segmentation):
