PAM50_t2 template contains signal outside spinal cord - impact on fMRI registration quality?

MRI Processing & SCT
1

Hi,

I’m analysing spinal MRI images for the first time and ran into something that might not be a problem at all but that I wanted to check to be sure.

SCT version: 7.0 OS: macOS

Pipeline:

  1. T1 processing: sct_register_to_template → sct_warp_template
  2. T2s processing: sct_register_multimodal (PAM50_t2 → T2s, initialised with T1 warp via -initwarp) → sct_warp_template (T2s space)
  3. fMRI processing: sct_register_multimodal using warped PAM50_t2.nii.gz as source image → sct_warp_template (fMRI space)

When viewing the original PAM50_t2 template, it contains substantial signal outside the spinal cord itself - vertebral bodies and surrounding tissue are clearly visible. After warping to subject T2s space, these surrounding structures are also present in the warped template.

In our fMRI pipeline, we use the warped PAM50_t2.nii.gz as the source image for registration to fMRI space:

bash:

sct_register_multimodal -i PAM50_t2.nii.gz \
                        -d fmri_mean.nii.gz \
                        -iseg PAM50_cord.nii.gz \
                        -dseg fmri_seg.nii.gz \
                        -param step=1,type=seg,algo=affine,slicewise=1,metric=MeanSquares,smooth=3:step=2,type=seg,algo=syn,metric=MeanSquares,smooth=1,iter=5

Questions

  1. Does the presence of surrounding anatomy in PAM50_t2.nii.gz negatively affect the quality of registration to fMRI space?

  2. Would it be better to mask PAM50_t2.nii.gz with the cord segmentation (PAM50_cord.nii.gz) before using it as a registration source, so that only cord signal is used for matching?

Warped_T2template
Warped_T2template1920×788 56.1 KB

Warped_T2templated_higherMinValue
Warped_T2templated_higherMinValue1920×1130 85.4 KB

2

Hi,

  1. Does the presence of surrounding anatomy in PAM50_t2.nii.gz negatively affect the quality of registration to fMRI space?

Given that you’re not using the image for registration (see type=seg in your sct_register_multimodal command), then no. The surrounding anatomy on the PAM50_t2 has no effect on the registration.

  1. Would it be better to mask PAM50_t2.nii.gz with the cord segmentation (PAM50_cord.nii.gz) before using it as a registration source, so that only cord signal is used for matching?

See previous comment. However, if you decide to use type=im, then it depends on the algorithm you’re using (bsplinesyn, syn, etc.). In general, it is good to only focus on the region close to the spinal cord, which can be done with a dilated mask inputted with flag -m