dtitk and ANTS for DTI spatial normalization

[Ping-Hong Yeh]

Hi Gary,

We recently found that the results of FA group comparisons using  DTITK for tensor warping and ANTS for FA warping are quite different on the same data set.  The significant clusters of group difference using ANTs were mainly found in the major fiber tracts and relatively symmetric on both hemispheres and , while the DTITK results were mainly in the gray matter and lateralization.  I'd assume to see similar results because the similarity of warping algorithms.?

Now I try to implement the Avants' 08 DTI & T1 multivariate method on the same data set for comparison.  In order to do that,  the tensor image needs to be upsampled to the T1 space, and then using ANTS algorithms for warping T1 & DTI  simultaneously.  Do you have any  suggestion on how to do this correctly using DTITK toolbox?

Thank you,

Ping

[Hui Zhang]

Hi Ping,

We recently found that the results of FA group comparisons using  DTITK for tensor warping and ANTS for FA warping are quite different on the same data set.  The significant clusters of group difference using ANTs were mainly found in the major fiber tracts and relatively symmetric on both hemispheres and , while the DTITK results were mainly in the gray matter and lateralization.  I'd assume to see similar results because the similarity of warping algorithms.?

DTI-TK is not meant to be used to analyze gray matter regions.  The metric used by DTI-TK is based on tensor and therefore it is very different from FA.  The DTI-TK's metric is designed to be particularly sensitive to aligning white matter but not gray matter.  Hence, when you do VBM type analysis, you need to apply the analysis only to the voxels of major white matter regions.

See this paper of mine that compares DTI-TK to ANTS:

http://www.nitrc.org/frs/download.php/1360/zhang07tmi.pdf

The similar results have been recently replicated in the following paper:

http://www.nitrc.org/forum/forum.php?forum_id=1892

The fluid method is very comparable to ANTS.

Now I try to implement the Avants' 08 DTI & T1 multivariate method on the same data set for comparison.  In order to do that,  the tensor image needs to be upsampled to the T1 space, and then using ANTS algorithms for warping T1 & DTI  simultaneously.  Do you have any  suggestion on how to do this correctly using DTITK toolbox?

You will need to first align DTI to T1 using scalar_affine_reg

scalar_affine_reg t1.nii.gz b0.nii.gz NMI 2 2 2 0.001

Notice that you need to use the b=0 image to align to the corresponding T1 image.

Next, you can use deformationSymTensor3DVolume to warp the DTI to the t1 space.

deformationSymTensor3DVolume -in dti.nii.gz -out dti_t1.nii.gz -trans b0.aff -target t1.nii.gz

The output dti_t1.nii.gz will be in the same space as your T1 image and with the same resolution.

Gary

[Ping-Hong Yeh]

Hi Gary, 

I got error while running the deformationSymTensor3DVolume

reading the buffer ... ** ERROR (nifti_image_read): failed to find header file for 'b0.aff'

Should the -trans of deformationSymTensor3DVolume only take the displacement field in nifti? 

The same command ran fine with affineSymTensor3DVolume though. 

Thanks. 

Ping

[Hui Zhang]

Sorry, I meant affineSymTensor3DVolume.  Indeed, the other one is for displacement only.

Gary