Real Data: first example

[oesteban]

I modify this ticket. The target is applying the code on one real case from @meribach 's pilot study.

[oesteban]

I'm applying it on my pilot, ts0.

1. Take DTI info out:

   cd /media/data/DTI_Repeatability/NIFTI/S001/ts0
   fslsplit DTI.nii.gz -t
   mv vol0000.nii.gz DTI_b0.nii.gz
   rm vol*.nii.gz

2. Extract brain

   bet DTI_b0.nii.gz DTI_b0_brain.nii.gz -f 0.3 -R

3. Registration (http://brainybehavior.com/neuroimaging/2011/04/using-freesurfers-bbregister-for-image-registration/):

   cd /media/data/DTI_Repeatability/FREESURFER
   export SUBJECTS_DIR=/media/data/DTI_Repeatability/FREESURFER; bbregister --s OE0_T1 --mov ../NIFTI/S001/ts0/DTI_b0_brain.nii.gz --reg ../NIFTI/S001/ts0/dti2t1.dat --dti --init-fsl

4. Check result

   tkregister2 --mov ../NIFTI/S001/ts0/DTI_b0_brain.nii.gz --reg ../NIFTI/S001/ts0/dti2t1.dat --surf

   

5. Transform T1 to dti space (and then check it, seems OK):

   mri_vol2vol --mov ../NIFTI/S001/ts0/DTI_b0.nii.gz --targ OE0_T1/mri/T1.mgz --reg ../NIFTI/S001/ts0/dti2t1.dat --o ../NIFTI/S001/ts0/T1-to-DTI.nii.gz --inv
   tkregister2 --mov ../NIFTI/S001/ts0/T1-to-DTI.nii.gz --targ ../NIFTI/S001/ts0/DTI_b0.nii.gz --reg ../NIFTI/S001/ts0/T1-to-DTI.nii.gz.reg

6. Transform surfaces to dti space:

   for hemi in lh rh; do for surf in pial white; do cmd="mri_surf2surf --s OE0_T1 --sval-xyz $surf --reg-inv dti2t1.dat DTI_b0_brain.nii.gz --tval ./$hemi.$surf.dti --tval-xyz --hemi $hemi"; echo $cmd; eval $cmd; done; done;

7. (Optional) Transform surfaces to vtk

   cd ../NIFTI/S001/ts0/
   mris_convert rh.pial.dti rh.pial.dti.vtk

8. Check that there are orientation problems with paraview:

   

[oesteban]

Updated methodology:

1. Take DTI info out:

   cd /media/data/DTI_Repeatability/NIFTI/S001/ts0
   fslsplit DTI.nii.gz -t
   mv vol0000.nii.gz DTI_b0.nii.gz
   rm vol*.nii.gz

2. Extract brain

   bet DTI_b0.nii.gz DTI_b0_brain.nii.gz -f 0.3 -R

3. Registration (http://brainybehavior.com/neuroimaging/2011/04/using-freesurfers-bbregister-for-image-registration/):

   cd /media/data/DTI_Repeatability/FREESURFER
   export SUBJECTS_DIR=/media/data/DTI_Repeatability/FREESURFER; bbregister --s OE0_T1 --mov ../NIFTI/S001/ts0/DTI_b0_brain.nii.gz --reg ../NIFTI/S001/ts0/dti2t1.dat --dti --init-fsl

4. Check result

   tkregister2 --mov ../NIFTI/S001/ts0/DTI_b0_brain.nii.gz --reg ../NIFTI/S001/ts0/dti2t1.dat --surf

5. Transform surfaces to dti space:

   for hemi in lh rh; do for surf in pial white; do cmd="python fsb_transform.py --targ DTI_b0_brain.nii.gz --orig ../FREESURFER/OE0_T1/surf/$hemi.$surf --o surf/test.$hemi.$surf --tfm bbreg/dti2t1.dat --paraview"; echo $cmd; eval $cmd; done; done;

6. Check results with Slicer:

   

[oesteban]

AND this is the result on the actual features (FA,MD):

FA:

MD:

I check that the test example is functional and close this issue :)

[zosso]

Question: are these results based on the proposed joint seg-reg algorithm? What exactly happens in the frontal part? is the brain cut (skull-strip artefact) or is this the distortion (to be corrected by the registration)?

[oesteban]

• No results yet, I hope to generate them tomorrow (there is still an issue with orientation, whereas 3D Slicer represents correctly the surfaces, paraview -and our tool- have an orientation problem).
• About the frontal part: this is, exactly, the effect of susceptibility distortion. The FA and MD maps are generated with dti reconstruction of a dMRI dataset after motion and eddy currents correction (again, without susceptibility).

I will compute the unwarped FA and MD using fieldmap correction, so you can see the performance.

On Mon, Apr 15, 2013 at 10:07 PM, zosso notifications@github.com wrote:

Question: are these results based on the proposed joint seg-reg algorithm? What exactly happens in the frontal part? is the brain cut (skull-strip artefact) or is this the distortion (to be corrected by the registration)?

— Reply to this email directly or view it on GitHubhttps://github.com/oesteban/ACWE-Registration/issues/53#issuecomment-16408247 .

[oesteban]

Finally, we have a 1st working example (blue=original contour, yellow=registered contour):

Conditions of the experiment:

• Only 1 surface (should work on the 4 surfaces -lh & rh, white & pial-).
• FA and MD parameters computed from synthetic examples (not very appropriate). I open #70 to discuss about this.
• Only 15 iterations.
• Parameters α and β from model configuration (obviously, if the intensity descriptors are not accurate, they will penalize too much the strong gradients).
• Deformation field parametrized on a grid of 16x16x8 nodes. This implies a physical spacing of 15.5mm (approx.).

I write these conditions because it might not seem a very good result, but actually it is. First thing to do is discussing #70 and initializing the problem better. Next step IMHO is implementing #7, given that it seems to provide a very isotropic transformation. Coupled or maybe at the same time, I will also face #49 .

Let me know your insights :)