Functional images registration

SCT
1

Dear SCT team,

We are trying to register the mean functional image to the PAM50 using 2 steps: func2T1w and T1w2PAM50. The second step looks good (see video t1w2pam50) but we have some issues with the first one (see video func2pam50):

  1. First when defining the destination file (sct_register_multimodal ):
  • If we use t2 we have the cord is enlarged.
  • If we use t2s, I notice that the first slices of the template (around C1 level) are missing and I have registration problem for the corresponding slices
  1. We also tested two different combinations of parameters and had trouble finding the best compromise:
  • params_func_to_T1w = ā€˜step=1,type=seg,algo=centermass,metric=MeanSquares:step=2,algo=bsplinesyn,type=seg,slicewise=1,iter=3ā€™
  • params_func_to_T1w=ā€˜step=1,type=seg,algo=slicereg,metric=MeanSquares:step=2,type=seg,algo=affine,metric=MeanSquares,gradStep=0.2:step=3,type=im,algo=syn,metric=CC,iter=5,shrink=2ā€™

Do you have any recommendation to improve our registration ?

Thank you !
Caroline

2

Hi @Caroline_Landelle,

My deepest apologies for the late response. I am writing to let you know that @jcohenadad is currently out of office, and will return on August 28th.

I will check in with other members of the lab to see if they can provide insight into your question. But, otherwise, I thank you for patience until Julien returns. :blush:

Kind regards,
Joshua

3

Hi Caroline, thank you for reaching out. I agree the T1wtoPAM50 registration looks good, and hopefully robust across your population. As for the GRE-EPI registration, itā€™s a bit hard for me to advise without looking at the data. If you could share 2-3 representative subjects, organised in BIDS format, Iā€™ll look for an optimal solution.

4

Hi Julien,

Here is the link with images from 2 participants:
https://mcgill-my.sharepoint.com/:f:/g/personal/caroline_landelle_mcgill_ca/EsWegBtFAPNFsaC9tOr4DoUBPJ6B4_pWq4cFEORcDEFhEg?e=j66Y0v

I hope that I put everything you need for your tests let me know if you need other images.

Thank you a lot for you help

5

Hi Caroline,

Sorry for the slow response. I am looking at your data now, but I donā€™t see any script that I can use to reproduce your results and perform further optimization. Could you please send me your script that I can use with sct_run_batch to process the two subjects you shared?

Thanks,
Julien

6

Hi Julien,

itā€™s my turn to reply late! Iā€™ve added a .sh script that can be run from the terminal + a readme file.
Iā€™m not familiar with the sct_run_batch script and it would take me some time to use it so I hope this format works with you!

https://mcgill-my.sharepoint.com/:f:/g/personal/caroline_landelle_mcgill_ca/EsWegBtFAPNFsaC9tOr4DoUBPJ6B4_pWq4cFEORcDEFhEg?e=gg0f48

Thanks,
Caroline

7

Hi Caroline,

Few comments:

  • The pipeline was not end-to-end but assumed some preprocessing were done before, e.g. segmentation of cord on T1w, registration between T1w and PAM50 template, etc. It is important to always provide a comprehensive pipeline script to make sure results are reproducible. I wrote the full pipeline (see below).
  • Your script required manual interventions (eg: ā€œSUBJECT=ā€œsub-A011ā€ # can be change by sub-006ā€), which is prone to human error. I made it fully automatic to run across a directory of subjects using sct_run_batch (see instructions below).
  • Your script did not include QC report generation. I strongly recommend using QC report generation to facilitate, systematize and make shareable the QC procedure, which is essential in any image analysis pipeline. Iā€™ve added QC report generation in the script below.
  • Your script registered funcā€“>T1w and then T1wā€“>PAM50. I recommend doing funcā€“>PAM50, using the T1wā€“>PAM50 intermediate warping field, to minimize the accumulation of registration errors. See new script.
  • The new script registers using C2 and C5 vertebral levels. Please change according to the regions you are most interested in (eg: C4-C7).
  • I used a mask (dilated spinal cord) to improve the performance of the PAM50->func registration
  • The data are not BIDS-compatible, making the analysis script ā€˜customā€™ and not shareable. For example, the suffix ā€œ_scā€ in an image file under anat/ is not authorized: sub-A006_T1w_sc.nii.gz. See BIDS specification and online BIDS validator.
  • I encourage you to watch the latest SCT course, especially the part on analysis pipeline.

Here is the script you can download: coreg_func_julien.sh (7.3 KB)

Open a Terminal, go where your folder ā€˜data/ā€™ is located and run (change YYYYMMDD_HHMMSS with whatever date/time you are running the pipeline):

sct_run_batch -script coreg_func_julien.sh -j -1 -path-data data -path-output results_YYYYMMDD_HHMMSS
Terminal output 
julien-macbook:~/Desktop/OneDrive_1_9-1-2023 $ sct_run_batch -script coreg_func_julien.sh -j -1 -path-data data -path-output results_20230913_121913

--
Spinal Cord Toolbox (git-master-76c27e16f13974de3ff8f81e25c92abe8f6d80ec)

sct_run_batch -script coreg_func_julien.sh -j -1 -path-data data -path-output results_20230913_121913
--

INFO SYSTEM
-----------
OS: osx (macOS-10.16-x86_64-i386-64bit)
CPU cores: Available: 12 | Threads used by ITK Programs: 1
RAM: Total 16384 MB | Available 7230 MB | Used 7912 MB

INPUT ARGUMENTS
---------------
batch_log: sct_run_batch_log.txt
config: null
continue_on_error: 1
email_from: null
email_host: smtp.gmail.com:587
email_to: null
exclude: null
exclude_list: null
ignore_ses: false
include: null
include_list: null
itk_threads: 1
jobs: -1
path_data: data
path_output: results_20230913_121913
path_segmanual: .
script: coreg_func_julien.sh
script_args: ''
subject_prefix: sub-
v: 1
zip: false

SCRIPT
------
git commit: ?!?
git origin: ?!?
Copying script to output folder...
/Users/julien/Desktop/OneDrive_1_9-1-2023/coreg_func_julien.sh -> /Users/julien/Desktop/OneDrive_1_9-1-2023/results_20230913_121913/coreg_func_julien.sh
Setting execute permissions for script file coreg_func_julien.sh ...

DATA
----
git commit: ?!?
git origin: ?!?

RUNNING
-------
Processing 2 subjects in parallel. (Worker processes used: 11).
Started at 12h19m14s: sub-A006. See log file /Users/julien/Desktop/OneDrive_1_9-1-2023/results_20230913_121913/log/coreg_func_julien_sub-A006.log
Started at 12h19m14s: sub-A011. See log file /Users/julien/Desktop/OneDrive_1_9-1-2023/results_20230913_121913/log/coreg_func_julien_sub-A011.log

Hooray! your batch completed successfully :-)

Started: 12h19m14s | Ended: 12h28m56s | Duration: 00h09m42s


To open the Quality Control (QC) report in a web-browser, type:
open /Users/julien/Desktop/OneDrive_1_9-1-2023/results_20230913_121913/qc/index.html

QC report (open index.html): qc.zip (2.2 MB)

Comments on the results:

  • Automatic vertebral labeling failed for sub-A006, so this should be corrected manually. See function label_if_does_not_exist in the script. You just need to create 2 points and place them under the derivatives/labels folder. Also see Analysis pipeline ā€” spine-generic documentation for inspiration about how to QC, manually correct, store corrections, and then use the corrections for re-running the analysis pipeline.
  • Cord segmentation for both subjects on the EPI scans need to be manually corrected (see previous point). This will positively impact the quality of the registration.

Hope that helps,
Julien

8

Dear Julien,
Thank you very much for taking the time to provide me an exemple and for these feedbacks,

  • Youā€™re right I didnā€™t follow the standards because my initial image covered both brain and spinal cord in this dataset and I have to play with different toolboxes (SPM, FSL, SCT) in a single preprocessing Python script. Iā€™ll try to improve the coding of the script for my next studies!
  • For the QC, I systematically check all images (brain + sc) with fsleyes after moco, segmentation, Coreg, denoising and also the tSNR improvement etcā€¦ (again, to have the same tool to examine QC for both brain and SC)
  • I have done the labelling of all vertebrae (Iā€™m interested in all cervical spinal levels)
  • A careful manual correction of the segmentation (T1w and func) was done before the coreg step.
  • Sorry, I had to change a little the BIDS format and include an additional level that I named ā€œstructureā€ to distinguish the brain or sc (spinal cord) after cropping my inital images in two. That is why I have the sub??_T1w_sc.nii.gz and the sub??_T1w_brain.nii.gz images for each individual. I hope to find a better standard in the future!
  • Thank you for pointing out the courses Iā€™m going to improve my brain and spinal cord pipeline and trying to find a standard that matches as much as possible both the brain and spinal cord communities!
  • Finally, your recommendation func > PAM50 using func>T1w intermediate warping field was a great solution! It improves my coregistration. I have done some tests and the coregistration looks a little better when Iā€™m using PAM50_t2s compared to PAM50_t2, except for the upper slices (z>960). In fact, these slices are missing in PAM50_t2s. Do you have a complete version of the PAm50_t2s image so I can test it on the entire cervical SC?

Thank you very much for your help !
Caroline

9

Unfortunately I donā€™t. The PAM50 was built with the T2w data and coverage available at that time. However the T2w is pretty close to the T2w in terms of contrast, so I recommend you use the T2w for registration if you need coverage above z=960

10

Dear Julien,
Thank you once more for your help.
I will then use the PAM50_t2 image

Caroline

1 Like