Data and code to test DT warping and reorientation in ANTs.
This code is updated for ANTs >= 2.6.2. See the reorient_tensor_image branch for older code.
A single subject scanned on a Siemens Prisma, with varying image orientations. The data is from
https://www.nitrc.org/plugins/mwiki/index.php/dcm2nii:MainPage#Diffusion_Tensor_Imaging
Please see the license file in rawData/license.txt if you want to use or redistribute this data.
The OASIS template is derived from the original by Nick Tustison, at
https://figshare.com/articles/ANTs_ANTsR_Brain_Templates/915436
The data was converted with dcm2niix, then renamed by series (axis, ortho, pitch, roll, yaw).
In other words, under nii/ we have:
axis: axis.bval axis.bvec axis.json axis.nii.gz
ortho: ortho.bval ortho.bvec ortho.json ortho.nii.gz
pitch: pitch.bval pitch.bvec pitch.json pitch.nii.gz
roll: roll.bval roll.bvec roll.json roll.nii.gz
yaw: yaw.bval yaw.bvec yaw.json yaw.nii.gz
The data is in the same physical space, but the acquisition planes vary.
Brain masks were created with FSL's bet.
The DT is computed with FSL (dtifit). It is then converted into ANTs format. No reorientation is applied at this stage, the components are just re-ordered into a NIFTI standard (lower triangular) file compatible with ITK I/O.
The scripts to do this are in nativeDT/, and can be run from that directory.
The registration is rigid. Deformable registration will be tested in a separate repository.
Alignment of principal directions is assessed by extracting the primary eigenvector of the tensor with ImageMath after registration / reorientation.
I have used fsleyes to visualize results. Note that FSL requires radiological orientation
for bvecs, regardless of whether or not the image index space is radiological. If
fslorient -getorient reports NEUROLOGICAL, fsleyes will automatically flip the tensor
vectors for display. You can toggle this in fsleyes settings.
ANTs attempts to handle this automatically by flipping the coordinate system of the tensors from NEUROLOGICAL images upon import and export. Inside ANTs, we require that the tensors match the index space of the image, such that they can be rebased to physical space by applying the header direction matrix.
The test data here is all RADIOLIGICAL, so this should not be an issue.
Example:
regScripts/reg.sh ortho roll
Or to run all registrations:
./runAll.sh
This aligns the FA images to each other, and to the OASIS T1 template. The rotation in physical space is very small, but getting the headers wrong can lead to large errors.
These tests mostly check that antsApplyTransforms is correctly reading and writing the deformed images.
For the OASIS registrations, the rigid alignment is only roughly correct, so it may be easier to view the warped FA image as the base layer rather than the T1.
Fixed images here are rotated with respect to the original FA images. They are designed to test the reorientation that occurs when the fixed and moving images are not aligned in physical space.
The fixed images for these experiments are in targetImagesWithRotation/.
To aid visual inspection, the rotations are quite large, so antsAI is called before antsRegistrationSyNQuick.sh.
fixed.nii.gz - the fixed image
movingToFixed_DTDeformed.nii.gz - DT resampled and reoriented into the fixed space.
movingToFixed_FADeformed.nii.gz - FA from movingToFixed_DTDeformed.nii.gz
The principal directions from each of these tensors is stored as a 4D vector, which FSL recognizes as a DT eigenvector. Eg,
movingToFixed_V1Deformed.nii.gz
ANTs normally writes these in NIFTI format, which is 5D with intent code NIFTI_INTENT_VECTOR. If you want this format, run
ImageMath 3 V1.nii.gz TensorToVector DT.nii.gz 2