Register AMU to PAM50 T2

[jcohenadad]

Related to #32

• if using the T2s and the T2 → we will likely run into issues with presence of visible GM inside the T2s
• if using the individual cord masks, we will run into issues because the edge of the cord is not defined similarly between the T2s and the T2
  • could be addressed by scaling (dilating) the existing cord mask on the T2w – eg: try different dilation values, then run registration, and visually decide what scaling factor works for the registration. A good starting point would be to apply a dilation factor on the T2w cord that would match the CSA of the AMU cord for a given level.

[jcohenadad]

I looked at running contrast-agnostic model on the PAM50 T2 and T1. Results look pretty good and consistent between the two contrasts (white: T1, red: T2):

Screen.Recording.2025-08-28.at.16.12.27.mov

[jcohenadad]

Also informative, although not surprising, is that the contrast-agnostic (in red) provides a slightly larger cord than the existing PAM50_cord file (in white):

Screen.Recording.2025-08-28.at.16.14.27.mov

[jcohenadad]

Here's a first pass:

cd $SCT_DIR/data/PAM50/template
# Create label on MNI-Poly-AMU to identify the z (top part of the visible template). The XY coordinate does not matter too much since pipeline will be followed by centermass registration.
sct_label_utils -i ~/Desktop/MNI-POLY-AMU/AMU15_T2star_sym.nii.gz -create 75,75,965,1 -o ~/Desktop/MNI-POLY-AMU/label_AMU15.nii.gz
# Create associated label in the PAM50
sct_label_utils -i PAM50_t2.nii.gz -create 70,70,959,1 -o label_PAM50.nii.gz
# Register AMU15 --> PAM50
sct_register_multimodal -i ~/Desktop/MNI-POLY-AMU/AMU15_T2star_sym.nii.gz -iseg ~/Desktop/MNI-POLY-AMU/AMU15_GW_sym.nii.gz -ilabel ~/Desktop/MNI-POLY-AMU/label_AMU15.nii.gz -d PAM50_t2.nii.gz -dseg PAM50_t2_seg.nii.gz -dlabel label_PAM50.nii.gz -param step=0,type=label,dof=Tx_Ty_Tz:step=1,type=seg,algo=centermass:step=2,type=seg,algo=bsplinesyn,metric=MeanSquares,iter=5,slicewise=1 -qc qc

Good first attempt, but requires further regularization along Z. Select slicewise=0 next time.

[jcohenadad]

oops!

Get image dimensions of destination image...
Loaded dest.nii (/private/var/folders/qd/2g7h7921023d_rzm472bf8jm0000gn/T/sct_2025-08-28_20-50-09_register-slicewise_30y43wwm/dest.nii) orientation RPI shape (141, 141, 991)
  matrix size: 141 x 141 x 991
  voxel size: 0.5mm x 0.5mm x 991mm --> ☠️

hum... maybe a glitch in the display of register multimodal? The file is fine:

julien-macbook:~ $ fslhd /private/var/folders/qd/2g7h7921023d_rzm472bf8jm0000gn/T/sct_2025-08-28_20-50-09_register-slicewise_30y43wwm/dest.nii 
filename	/private/var/folders/qd/2g7h7921023d_rzm472bf8jm0000gn/T/sct_2025-08-28_20-50-09_register-slicewise_30y43wwm/dest.nii

sizeof_hdr	348
data_type	UINT8
dim0		3
dim1		141
dim2		141
dim3		991
dim4		1
dim5		1
dim6		1
dim7		1
vox_units	mm
time_units	Unknown
datatype	2
nbyper		1
bitpix		8
pixdim0		-1.000000
pixdim1		0.500000
pixdim2		0.500000
pixdim3		0.500000

Issue opened about the glitch spinalcordtoolbox/spinalcordtoolbox#5021

[jcohenadad]

Much better with slicewise=0, but we still notice slight discontinuities along the Z axis:

sct_register_multimodal -i ~/Desktop/MNI-POLY-AMU/AMU15_T2star_sym.nii.gz -iseg ~/Desktop/MNI-POLY-AMU/AMU15_GW_sym.nii.gz -ilabel ~/Desktop/MNI-POLY-AMU/label_AMU15.nii.gz -d PAM50_t2.nii.gz -dseg PAM50_t2_seg.nii.gz -dlabel label_PAM50.nii.gz -param step=0,type=label,dof=Tx_Ty_Tz:step=1,type=seg,algo=centermass:step=2,type=seg,algo=bsplinesyn,metric=MeanSquares,iter=5,slicewise=0 -qc qc

Screen.Recording.2025-08-29.at.09.15.43.mov

[jcohenadad]

Slicereg helps a little bit, but at other places we still notice some XY shifts

sct_register_multimodal -i ~/Desktop/MNI-POLY-AMU/AMU15_T2star_sym.nii.gz -iseg ~/Desktop/MNI-POLY-AMU/AMU15_GW_sym.nii.gz -ilabel ~/Desktop/MNI-POLY-AMU/label_AMU15.nii.gz -d PAM50_t2.nii.gz -dseg PAM50_t2_seg.nii.gz -dlabel label_PAM50.nii.gz -param step=0,type=label,dof=Tx_Ty_Tz:step=1,type=seg,algo=slicereg,poly=5:step=2,type=seg,algo=bsplinesyn,metric=MeanSquares,iter=5,slicewise=0 -qc qc

[jcohenadad]

Skipping the bsplinesyn registration solves this XY shift. The registration is much better now.

sct_register_multimodal -i ~/Desktop/MNI-POLY-AMU/AMU15_T2star_sym.nii.gz -iseg ~/Desktop/MNI-POLY-AMU/AMU15_GW_sym.nii.gz -ilabel ~/Desktop/MNI-POLY-AMU/label_AMU15.nii.gz -d PAM50_t2.nii.gz -dseg PAM50_t2_seg.nii.gz -dlabel label_PAM50.nii.gz -param step=0,type=label,dof=Tx_Ty_Tz:step=1,type=seg,algo=slicereg,poly=2 -qc qc

qc.zip

[jcohenadad]

Fixed in 298a558