Funky thoracic warping

Hello SCT developers:

I am warping some thoracic data to the SCT, and just observed some very funky warping. Has anybody seen this before? I have never observed this previously in cervical data. I am using the current version of the SCT (via git pull as of today - Dec 3, 2019). Thank you!


hi Rob,
yes, it could be due to different things. Maybe the easiest would be to share with me your data + syntax that you ran so i can debug locally

1 Like

Thanks, Julien. I just shared these data with you. Also, this was the first of several data sets that were processed in the same way, and this streaking artifact was present in all but one of the other data sets. -Rob

hi @barryrl,

a few things:

  • PAM50 template has maximum disc value of 21 (you included label 22, which does not exist)
  • When specifying more than 2 labels, the algorithm is estimating a non-linear z-transformation to perfectly align each disc. The downside is that above and below the edge discs (in your case: 15 and 21), there is no warping field and the image looks “funky”. To prevent that issue, you can either provide more discs (although in your case you reached the “21 limit”), or use two discs maximum (in which case a z-affine transformation is estimated, and the warping field covers the entire S-I axis).

Here is the syntax:

# Remove superfluous discs
sct_label_utils -i TL_labels.nii.gz -remove 16,17,18,19,20,22 -o TL_labels_15-21.nii.gz 
# Run registration
sct_register_to_template -i t2.nii -s t2_seg.nii -ldisc TL_labels_15-21.nii.gz -c t2 -qc qc

Here is the qc report: (2.2 MB)


Brilliant! The use of only two labels works across all TL data sets. Thank you! -Rob

1 Like