After processing dozens of subjects through tracula I have found some subjects with tracts that are one voxel along the entire tract, for example the forceps major. After reviewing the data I believe the bvecs files are pointing the correct direction, I believe the registration between the T1/freesurfer segmentation and DTI is reasonably good (registration was performed with bbr), at least the anat_brain_mask-vent.bbr and dtifit_FA are reasonable and cortex.bbr and lowb_brain seem reasonably well registered. The brain extraction does look poor in that a large chunk of posterior skull remains in the anat_brain_mask-vent.bbr image but the as the registrations, segmentations and parcellations look OK I believe this isn't the cause of my problems.
I do see one issue: the registration between the dtifit_FA.bbr and fmajor_PP_avg33_mni_bbr_logprior_0 images. Sections of the logprior volume are entirely in the lateral ventricles of the DTI data. Please see attached jpg. This subject large lateral ventricles and I am wondering if such subjects can be run successfully in tracula. Much thanks for any suggestions.
Hi Bryon - It does look like this individual and the tract atlas are not in good alignment. The one step that requires good alignment is for choosing an initial guess for the pathway. If the fmajor from all or most of the subjects that are included in the tract atlas goes through the ventriclesi of your subject, it'll be tough to get a good initialization so reconstruction will fail.
You can try the reinit option but in this case you'd have to try it many times since pretty much all the atlas samples go through your subject's ventricles.
The last resort for such a subject is to initialize it manually. If you upload this example subject for me here I can try to see what can be done: https://www.nmr.mgh.harvard.edu/facility/filedrop/index.html
In principle, once the initialization is ok, the actual path reconstruction should work even with poor spatial alignment, since it only uses information on the neighboring aparc+aseg structures and not on the spatial location of the path. This'll be a chance to test this hypothesis!
Hope this helps, a.y
On Thu, 29 Nov 2012, bryon mueller wrote:
After processing dozens of subjects through tracula I have found some subjects with tracts that are one voxel along the entire tract, for example the forceps major. After reviewing the data I believe the bvecs files are pointing the correct direction, I believe the registration between the T1/freesurfer segmentation and DTI is reasonably good (registration was performed with bbr), at least the anat_brain_mask-vent.bbr and dtifit_FA are reasonable and cortex.bbr and lowb_brain seem reasonably well registered. The brain extraction does look poor in that a large chunk of posterior skull remains in the anat_brain_mask-vent.bbr image but the as the registrations, segmentations and parcellations look OK I believe this isn't the cause of my problems.
I do see one issue: the registration between the dtifit_FA.bbr and fmajor_PP_avg33_mni_bbr_logprior_0 images. Sections of the logprior volume are entirely in the lateral ventricles of the DTI data. Please see attached jpg. This subject large lateral ventricles and I am wondering if such subjects can be run successfully in tracula. Much thanks for any suggestions.
Hi Bryon - I was able to reconstruct the forceps major in your subject after tweaking some of the 7 control points for the initialization. I'm attaching the text file with the corrected control points - just drop it in dlabel/diff and rerun the path reconstruction (don't use reinit, though!)
I'd definitely not go below 7 control points for this subject b/c of how convoluted the shape of the forceps major is (due to enlarged ventricles & thin corpus callosum). You might've even gotten it to work automatically if you'd gone to more than 7 control points but I haven't tried.
I'll try to incorporate some of this logic into the program in the next version but if you need to do this edit on more cases in the meantime:
1. Fit a spline to your control points: dmri_spline --cpts dlabel/diff/fmajor_PP_avg33_mni_bbr_cpts_7.txt \ --mask dmri/dtifit_FA.nii.gz \ --out dlabel/diff/fmajor_PP_avg33_mni_bbr_cpts_7.nii.gz
2. Display dlabel/diff/fmajor_PP_avg33_mni_bbr_cpts_7.nii.gz on top of dmri/dtifit_FA.nii.gz and see where the spline strays off the white matter.
3. Edit dlabel/diff/fmajor_PP_avg33_mni_bbr_cpts_7.txt by changing one of the control points that are near the part that's off the white matter.
4. Repeat.
Sorry, this is so not user friendly but we're working on improvements!
Hope this helps a bit, a.y
On Thu, 29 Nov 2012, bryon mueller wrote:
After processing dozens of subjects through tracula I have found some subjects with tracts that are one voxel along the entire tract, for example the forceps major. After reviewing the data I believe the bvecs files are pointing the correct direction, I believe the registration between the T1/freesurfer segmentation and DTI is reasonably good (registration was performed with bbr), at least the anat_brain_mask-vent.bbr and dtifit_FA are reasonable and cortex.bbr and lowb_brain seem reasonably well registered. The brain extraction does look poor in that a large chunk of posterior skull remains in the anat_brain_mask-vent.bbr image but the as the registrations, segmentations and parcellations look OK I believe this isn't the cause of my problems.
I do see one issue: the registration between the dtifit_FA.bbr and fmajor_PP_avg33_mni_bbr_logprior_0 images. Sections of the logprior volume are entirely in the lateral ventricles of the DTI data. Please see attached jpg. This subject large lateral ventricles and I am wondering if such subjects can be run successfully in tracula. Much thanks for any suggestions.
freesurfer@nmr.mgh.harvard.edu
-
Anastasia Yendiki -
bryon mueller