Binary mask-based methods or Atlas-based methods

SCT
1

Dear SCT Team:
I have a question about Binary mask-based methods and Atlas-based methods.
Which of the following two methods provides more reliable measurements?

  1. Measuring MTR using GM and WM segmentation created from T2* data
  2. Measuring MTR using white matter and grey matter atlases included in the PAM50 template
2

Hi @ryu,

If the GM segmentation is reliable (which depends on the image quality), I would say option 1. If not, option 2.

Cheers,
Julien

4

The CNR between WM and GM is borderline, I would say. For example, GM is improperly segmented on a few slices, example below:

Screenshot 2025-02-07 at 8.28.18 AM
Screenshot 2025-02-07 at 8.28.18 AM1470Ɨ1054 268 KB

Also, maybe not all subjects in your cohort will have the same quality.

So my conclusion is that relying on the PAM50 WM/GM atlas is probably more reliable.

5

@nlaines Could you please try your new contrast-agnostic GM segmentation model? Iā€™m curious to see how it compares to the existing method (maybe share the output QC report).

6

Hello @ryu

Sure! @jcohenadad

Here is the QC:
qc.zip (439.5 KB)
QC Legend:

  1. t2s_gmseg.nii.gz ā†’ from sct_deepseg_gm
  2. t2s_seg_gm_contrast_agnostic.nii.gz ā†’ from seg_gm_contrast_agnostic model version: r20250204

and here the .nii.gz mask:
t2s_seg_gm_contrast_agnostic.nii.gz (18.7 KB)

Best,

Nilser

1 Like
7

Nice! Your method works better :rocket:

@ryu you might want to switch to it if you end up choosing option 1

1 Like
8

Dear @nlaines @jcohenadad

Thank you very much.

Please tell me the command used to create t2s_seg_gm_contrast_agnostic.nii.gz.

9

Hi @ryu
The model seg_gm_contrast_agnostic is currently implemented in my SCT branch nlm/add_gm_contrast_agnostic_model , commit 311307e24ae4f9bebd98574569294ab93f45ebd3
if you are working on a development SCT version, you just have to change to my branch and then you can install the model using:

sct_deepseg -install-task seg_gm_contrast_agnostic

and run :

sct_deepseg -task seg_gm_contrast_agnostic -i t2s.nii.gz

Best,

Nilser

11

Looks good!