ANTS registration help

Summary of what happened:

I am trying to register a T2w image (0.2mm, 1mm, 0.2mm) to a T2w SBREF image (0.8mm isotropic). Both are partial field of view. The initial guess overlay looks okay but the final output of the below command has the T2w image quite drastically shifted. The SBREF has quite a lot of detail so I thought that registering these two images together should in theory be okay.

Do you have any recommendations for improving this registration? I am quite new to ANTs and appreciate any help. Thank you!

Command used (and if a helper script was used, a link to the helper script or the command generated):

Generate an initial guess overlay using identity transform

antsApplyTransforms -d 3 -i {T2_image} -r {sbref_0p8mm} -o {initial_guess_output} -n Linear --transform [identity]


# Step 2: Register the resampled T2 image to the sbref 0.8mm image
antsRegistration --dimensionality 3 \\
                 --float 0 \\
                 --output [{outputTransformPrefix},{warpedImage}] \\
                 --interpolation BSpline[3] \\
                 --use-histogram-matching 1 \\
                 --initial-moving-transform [{T2_image},{sbref_0p8mm},1] \\
                 --transform Affine[0.1] \\
                 --metric MI[{T2_image},{sbref_0p8mm},1,32] \\
                 --convergence [1000x500x250x100,1e-6,10] \\
                 --shrink-factors 8x4x2x1 \\
                 --smoothing-sigmas 3x2x1x0vox>

Version:

ants_2.4.3

Screenshots / relevant information:

SBREF image

T2w image

Initial overlay (see code above)

Final result

Hi everyone,

Just an update incase anyone has a similar problem. I solved this by doing the following:

T2w pFOV 0.2, 1, 0.2mm → T1w whole-brain 1mm iso using antsRegistration.
SBREF pFOV 0.8mm iso → T1w whole-brain 1mm iso with fsls epi_reg and a wm parcellation from FAST.
Inverting this transformation matrix and applying it to the transformed T2w image.

2 Likes