I would agree that this suggests a FreeSurfer issue.  Usually it tries to avoid 
crossing the midline, but for some reason has still done this in this 
particular case.

Peace,

Matt.

From: 
<[email protected]<mailto:[email protected]>>
 on behalf of Timothy Coalson <[email protected]<mailto:[email protected]>>
Date: Monday, November 7, 2016 at 5:49 PM
To: Simon Baker <[email protected]<mailto:[email protected]>>
Cc: "[email protected]<mailto:[email protected]>" 
<[email protected]<mailto:[email protected]>>
Subject: Re: [HCP-Users] Custom parcellation for HCP data

I think someone else will have to answer about whether that is within expected 
error.  The surfaces themselves cross over each other in those captures, so I 
think freesurfer just couldn't decide where the cortex ended.  Any idea how 
often this happens (how many subjects, how much overlap)?

Tim



On Mon, Nov 7, 2016 at 4:25 PM, Simon Baker 
<[email protected]<mailto:[email protected]>> wrote:
Hi Tim,

Have you had a chance to look into this issue?

Thanks,

Simon


On 18 October 2016 at 15:15, Simon Baker 
<[email protected]<mailto:[email protected]>> wrote:
Hi Tim,

Using FreeView, I loaded HCP subject 101410's MNINonLinear/T1w_restore_brain, 
MNINonLinear/aparc+aseg, and MNINonLinear/aparc.a2009s+aseg volumes, as well as 
our custom parcellation of the right hemisphere (custom_R) and left hemisphere 
(custom_L). Then I loaded the following surfaces: 
MNINonLinear/101410.R.pial.164k_fs_LR.surf.gii; 
MNINonLinear/101410.L.pial.164k_fs_LR.surf.gii. After that, I located a voxel 
near to where the right and left pial surfaces overlap: [130, 222, 92] (this is 
the location of the cursor in the attached SCREENSHOTS.pdf). The first row of 
data in the below table shows the value of this voxel in each parcellation.

coordinates    aparc+aseg    aparc.a2009s+aseg    custom_R    custom_L
[130, 222, 92]    2026    12106    2069    1094
[131, 222, 92]    2026    12106    0    1094
[132, 222, 92]    2026    12106    0    1094
[133, 222, 92]    2026    12106    0    1094
[134, 222, 92]    1026    11106    0    1094

As you can see, this voxel is labelled as 2026 in aparc+aseg and 12106 in 
aparc.a2009s+aseg. You can also see that this voxel is labelled 2069 in 
custom_R and 1094 in custom_L, indicating that it has been included in our 
custom parcellation of the right hemisphere as well as our custom parcellation 
of the left hemisphere. This is an example of overlap issue we are concerned 
about.

When we move to a neighbouring voxel (i.e., [131, 222, 92]), which is one voxel 
to the left of the starting voxel and within an area enclosed by the left pial 
surface (not the right pial surface), the value remains unchanged in all 
parcellations except custom_R, wherein the voxel is labelled as 0. This pattern 
is repeated as we move to a 2nd voxel (i.e., [132, 222, 92]) and a 3rd voxel 
(i.e., [133, 222, 92]) to the left of the starting voxel. It is not until we 
move to a 4th voxel (i.e., [134, 222, 92]) to the left of the starting voxel 
that the voxel is universally labelled as part of a structure in the left 
hemisphere: 1026 in aparc+aseg, 11106 in aparc.a2009s+aseg, 0 in custom_R, and 
1094 in custom_L.

Why are voxels that are clearly outside the pial surface of the right 
hemisphere labelled as part of a structure in the right hemisphere? Is there a 
problem with the aparc+aseg and aparc.a2009s+aseg parcellations in this regard? 
Or does the apparent mislabelling fall within an acceptable margin for error? 
And if so, is it possible that the accuracy of our custom parcellation is 
acceptable, despite the fact there is some overlap between the right hemisphere 
and left hemisphere?

Simon Baker
Brain & Mental Health Laboratory
Monash Institute of Cognitive & Clinical Neurosciences
Monash University


On 5 October 2016 at 07:56, Timothy Coalson 
<[email protected]<mailto:[email protected]>> wrote:
You might want to compare them to the T1 in an example subject (and perhaps 
also volume surface outline for pial), to see what is really going on.  The 
pial surfaces of the two hemispheres shouldn't overlap outside the medial wall.

Are you mapping to the structural volume resolution, or to the fMRI resolution? 
 Larger voxels makes this problem harder to resolve.  The label to volume 
mapping is also a bit greedy, as partial-volumed voxels are labeled 
unconditionally (the "unlabeled" value can win out, if the data from the label 
file suggests more volume in a voxel for "unlabeled", but the partial volume 
that lies outside the surfaces is ignored, even if it dominates).  As such, if 
it is only a few voxels, and only the very edges, it may not really matter how 
you resolve the overlap.

Tim


On Tue, Oct 4, 2016 at 12:44 AM, Simon Baker 
<[email protected]<mailto:[email protected]>> wrote:
Hi Tim,

Thanks for your email. To clarify, when I said voxels on the medial wall, I was 
referring to voxels near the sagittal midline of the brain (i.e., I wasn't 
referring to voxels that have been labelled as medial wall). As such, these 
voxels have been labelled as part of a structure in the left hemisphere and 
also labelled as part of a structure in the right hemisphere. In which case, 
how do you suggest we deal with the overlap issue?

Thanks again,

Simon


On 4 October 2016 at 13:02, Timothy Coalson 
<[email protected]<mailto:[email protected]>> wrote:
Note that this command just makes the left hemisphere take precedence - you 
said the overlap was in medial wall, which is more of an artifact of processing 
than a structure of importance, so this should be fine.  The surfaces shouldn't 
have any overlap on real structures, so ribbon constrained surface to volume 
mapping should prevent any other serious overlaps.

Tim


On Mon, Oct 3, 2016 at 8:56 PM, Timothy Coalson 
<[email protected]<mailto:[email protected]>> wrote:
Assuming that your label values don't have any overlap (each integer uniquely 
identifies not only the area but also the hemisphere), you can do the math part 
with wb_command -volume-math:

wb_command -volume-math 'L + (L == 0) * R'  ${Subject}.custom_raw.nii.gz -var L 
${Subject}.L.custom.nii -var R ${Subject}.R.custom.nii

You can then make the combined label names text file and use it to reimport the 
label file:

wb_command -volume-label-export-table ${Subject}.L.custom.nii 1 
${Subject}.L.custom.txt
wb_command -volume-label-export-table ${Subject}.R.custom.nii 1 
${Subject}.R.custom.txt
cat ${Subject}.L.custom.txt ${Subject}.R.custom.txt > ${Subject}.custom.txt
wb_command -volume-label-import ${Subject}.custom_raw.nii.gz 
${Subject}.custom.txt ${Subject}.custom.nii.gz

I also suggest using .nii.gz, label files compress very well.  Unlike FSL, 
workbench doesn't have an environment variable controlling the output format, 
you must specify full filenames.

Tim


On Mon, Oct 3, 2016 at 8:29 PM, Simon Baker 
<[email protected]<mailto:[email protected]>> wrote:
Hi all,

We want to use our own custom parcellation with the connectome project data. 
Ultimately the parcellation needs to be a NIFTI volume.

We mapped our own custom parcellation from fsaverage (.annot file in FreeSurfer 
space) onto an HCP subject (GIFTI surface .label.gii file in 
${Subject}/MNINonLinear space) separately for each hemisphere, and we would 
like to convert the GIFTI surface files into a NIFTI volume file. Although this 
can be done with the following commands . . .

[MNINonLinear]$ wb_command -label-to-volume-mapping 
${Subject}.L.custom.164k_fs_LR.label.gii 
${Subject}.L.sphere.164k_fs_LR.surf.gii T1w.nii.gz ${Subject}.L.custom.nii 
-ribbon-constrained ${Subject}.L.white.164k_fs_LR.surf.gii 
${Subject}.L.pial.164k_fs_LR.surf.gii

[MNINonLinear]$ wb_command -label-to-volume-mapping 
${Subject}.R.custom.164k_fs_LR.label.gii 
${Subject}.R.sphere.164k_fs_LR.surf.gii T1w.nii.gz ${Subject}.R.custom.nii 
-ribbon-constrained ${Subject}.R.white.164k_fs_LR.surf.gii 
${Subject}.R.pial.164k_fs_LR.surf.gii

[MNINonLinear]$ fslmaths ${Subject}.L.custom.nii -add ${Subject}.R.custom.nii 
${Subject}.custom.nii

. . . we have found that some voxels on the medial wall are labelled as both a 
left hemisphere region and a right hemisphere region.

Using wb_command, how can we generate a NIFTI volume file containing both the 
left hemisphere parcellation and the right hemisphere parcellation and ensure 
that each voxel is labelled as only one region?

Kind regards,

Simon Baker
Brain & Mental Health Laboratory
Monash Institute of Cognitive & Clinical Neurosciences
Monash University


_______________________________________________
HCP-Users mailing list
[email protected]<mailto:[email protected]>
http://lists.humanconnectome.org/mailman/listinfo/hcp-users








_______________________________________________
HCP-Users mailing list
[email protected]<mailto:[email protected]>
http://lists.humanconnectome.org/mailman/listinfo/hcp-users

________________________________
The materials in this message are private and may contain Protected Healthcare 
Information or other information of a sensitive nature. If you are not the 
intended recipient, be advised that any unauthorized use, disclosure, copying 
or the taking of any action in reliance on the contents of this information is 
strictly prohibited. If you have received this email in error, please 
immediately notify the sender via telephone or return mail.

_______________________________________________
HCP-Users mailing list
[email protected]
http://lists.humanconnectome.org/mailman/listinfo/hcp-users

Reply via email to