Register the other shim methods to the GRE CoV scan

[jcohenadad]

No description provided.

[dpapp86]

Since sub-04 is the one with the worst image quality (chunky lad), I propose using it as a benchmark.

Likewise, patient or volume tend to have the lowest signal intensity/contrast, so they are a good worst case scenario. For sub-04, I am looking at CoV-2-CoV (good sanity check), CP-2-CoV, Volume-2-CoV, and Target-2-CoV. I looked at a one-step coreg using affine, slicereg and bsplinesyn. And dl for good measure.

Code in question:
for subject in subjects: os.chdir(os.path.join(path_data, subject, "anat")) if subject == 'sub-04': for shim_mode in shim_modes: !sct_register_multimodal -i {subject}_acq-{shim_mode}_T2starw.nii.gz -d {subject}_acq-CoV_T2starw_crop.nii.gz -dseg ../anat/{subject}_acq-CoV_T2starw_crop_seg.nii.gz -param step=1,type=im,algo=slicereg,metric=CC -qc {path_qc}

Using affine:

Quite bad on patient, see jumping SC at cursor:

Screen.Recording.2024-01-11.at.22.15.17.mov

Using slicereg:
Same jumping:

Screen.Recording.2024-01-11.at.22.27.07.mov

Using bsplinesyn:
Very poor

Screen.Recording.2024-01-11.at.22.41.41.mov

Using DL:
Best one so far

Screen.Recording.2024-01-11.at.22.49.55.mov

What is your opinion @jcohenadad ?

[jcohenadad]

Thank you for these investigations. The DL algo definitely looks promising.

I've had hard time doing a proper registration on sub-04, see preliminary investigations in #57 (comment).

The issue is partly because of #58.

However, given that we don't absolutely need to report GRE signal across all shim modes (I think there is enough material in the paper so far), I would skip that step for now.