DTI metrics per-level

Hello team,

I am trying to find the DTI metrics for all levels, but it usually generates none for levels 1, 6 and 7. Since level 6 and 7 are important for us, I wanted to ask whether there is a way to have the DTI metrics (FA, MD, RD) for all levels or not?


Hi Maryam,

This is possible, but requires the following:

  • Registering the PAM50 template to your DTI scans
  • Ensuring that there is sufficient overlap between the the registered PAM50 template and the regions of interest for your DTI scans.
  • And finally, using the -perlevel option combined with specifying the correct vertebral levels using the -vert option for the sct_extract_metric command

To double-check these details, then, could you please share:

  • A screenshot of a sagittal view of the label/template/PAM50_levels.nii.gz file (the one that has been warped using sct_warp_template) overlaid on top of your DTI scan
  • The exact commands you have used in the processing pipeline (e.g. registration, template warping, metric extraction, etc.)

Thank you kindly,

1 Like

Thanks @joshuacwnewton

Okay, I will follow your points while I could access to my system.
But as a quick question, I was wondering whether “restore” method could help in this situation or not (obtaining values for all levels and accurate enough).

P.S. I attached the commands used for the processing.

dwi-metrics.sh (2.6 KB)

Thanks very much

the “restore” method would likely not help in this situation. The “restore” method is just a different way to fit the diffusion tensor (using robust fitting).

1 Like

Thank you Julien. So, how to generate the DTI metrics for all levels? I could find the required metrics for all levels in anatomical image, but DWI usually misses some levels (1, 6, 7).

@joshuacwnewton suggested a few QC steps here: DTI metrics per-level - #2 by joshuacwnewton

we need you to do this in order to be able to properly help you

1 Like

Here are the pam50-levels overlaid on DTI (as you observe they are not correctly mapped – in the end metrics are calculated only for levels 1 to 4, which does not seem to be corresponding to the true level according this figure.).


Also I added an screenshot of labelling done for T2w image, to show that labelling is done correctly for anatomical image.


The bash file used for analysis of DWI image, has been attached in previous replies.

Thank you very much.

in the end metrics are calculated only for levels 1 to 4, which does not seem to be corresponding to the true level according this figure.).

Actually I would say the opposite: this figure shows that the vertebral levels covered by the DWI scan are C1 to C4, so it is normal that C5 and below give you ‘none’ DTI metrics values.

Now, in order to make sure that the patient did not move (or was repositioned) between the anatomical image and the DWI image, I suggest you show an overlay of the DWI scan on the T2w anatomical image.

1 Like


you’re right… since I need to extract DTI metrics from other levels (5, 6, 7) as well, what can I do to include them?

You need to acquire DWI data at levels 5,6,7

1 Like

You’re right… I understood, and checked some more images. It seems that the no. of vertebral levels covered by the DWI scan are usually lesser than 7 (whole cervical cord). As the last question, how could I make sure that the DTI metrics extracted in available levels are accurate? Thank you Julien.

By ensuring that (this is a non exhaustive list):

  • The acquisition has sufficient SNR
  • The level of artifact (eg susceptibility) is reasonably low
  • The amount of volume-to-volume motion is reasonably low and properly corrected (look at the QC report)
  • The segmentation is properly done (see QC)
  • The registration to the PAM50 template is properly done (see QC)
  • The mask within which your metrics are extracted includes sufficient voxels. See notably (sct_extract_metric) what is Total Fractional Volume? - #2 by jcohenadad

This list is not specific to your question, but includes basic things to watch out for when doing medical image analysis.

1 Like