On Sat, May 20, 2017 at 4:12 PM, John Griffiths <j.davidgriffi...@gmail.com> wrote:
> > Hi Tim. > > Thanks for the detailed response. > > So you're saying group-representative coordinates are a difficult issue. > > For present purposes I would be ok with just individual subject native > space coordinates from this parcellation. > > I'm still not clear how to get these. > > > By way of context, I have a set of ROIs from HCP resting fmri data that I > obtained using > > wb_command -cifti-parcellate rfMRI_REST1_LR_Atlas.dtseries.nii Gordon333_ > FreesurferSubcortical.32k_fs_LR.dlabel.nii COLUMN parc_ts.ptseries.nii > > which gives me 352 time series. > > What I'm after now are the coordinates of the vertices/voxels (and ideally > also text labels) that contributed to each of these time series ROIs. > Individual subject coordinates work fine, whether native or MNI space. It sounds like you want to know which vertices are in each area, and the simplest way to get there is to use -cifti-separate to generate left and right hemisphere gifti label files (use -label, and extension .label.gii) - the arrays that these files represent have the same number of elements as there are surface vertices, and they are in the same order. Use -label-export-table or -file-information to get a listing of the label names and their key values (and rgba colors). For instance, if the label table says that "R_V2" is key "4", then every element in the array in the gifti label file that equals 4 is part of "R_V2". The coordinates are in surface files (.surf.gii), along with the connectedness information (triangles). If this poses a challenge to load and examine the coordinates with whatever tool you want to use, you can use -surface-coordinates-to-metric to convert just the coordinates to a 3-column .func.gii file. > The BALSA repository that I got the parcellation from included amongst > other things the following files > > Gordon333_FreesurferSubcortical.32k_fs_LR.dlabel.nii > S900.L.inflated_MSMAll.32k_fs_LR.surf.gii > S900.R.inflated_MSMAll.32k_fs_LR.surf.gii > > The dlabel file is size 1x91282 > > The surf gii files contain surfaces of with 32492 > > Am I to assume / supposed to know that something like the first 32492 > elements of the dlabel file > correspond to the left hemisphere surface, the second 32492 to the right > hemisphere surface, and > the remaining to the subcortical voxels? > CIFTI files do not have a section that maps 1-to-1 to vertices, they have the ability to exclude vertices from a surface, which we use to exclude the medial wall (our standard cifti space has about 29,000 vertices of each hemisphere represented in the file, out of 32,492 vertices in the matching surfaces). The -cifti-separate command takes care of this for you. See wb_command -cifti-help for more information. > But actually I know this can't be the case because in the first 32492 > elements of the dlabel file there are > 194 unique values, and in the second 32492 elements there are 174 unique > values. That said, the first > two blocks have lots of shared entries, so the total number of unique > labels does sum up to the correct > number, more or less. But many are in both the first and second blocks. > You did get fairly close to the actual layout, with ~3000 vertices being treated as the wrong hemisphere, and ~6000 subcortical voxels being treated as part of one cortical hemisphere. If most of the values are reused, then maybe this parcellation didn't use separate labels for left and right hemisphere? We have a way to fix this, and I thought we used it to try to release only files that split the labels by hemisphere. > So I'm stumped. > > > Please advise. > > > Much appreciated, > > > john > > > > > > > > > On 20 May 2017 at 15:01, Timothy Coalson <tsc...@mst.edu> wrote: > >> You can get the subcortical parts of that label file using >> -cifti-separate with the -volume-all option, which outputs a nifti volume. >> >> As for the cortical data, no the problem isn't simple. The cortical >> surface data doesn't really have meaningful "group coordinates" the way we >> see it, because volume registration doesn't fully align cortex across >> subjects. If this were the HCP MMP 1.0, conversion to volume would be even >> more problematic, since it relies on registration based on areal features, >> and not folding patterns. This is why we recommend that analysis of cortex >> data, especially group analysis, be done on surfaces (where intersubject >> correspondences don't rely on anatomical coordinates), rather than in the >> volume. >> >> If you still want to translate the cortical data to a group volume space, >> be aware that group average surfaces are not suitable to get to that volume >> space, because the averaging across subjects of cortical coordinates causes >> significant amounts of smoothing, reducing the amount of folding >> information in the group average surface. This causes the average surfaces >> to slice through white matter and csf in regions with high folding and high >> subject variability. >> >> The problem of wanting group average cortical data in the volume doesn't >> have a particularly good solution. Using individual surfaces from many >> subjects to map the data to volume will at least give an honest answer, >> though probably disappointing. >> >> Tim >> >> >> On May 20, 2017 11:49 AM, "John Griffiths" <j.davidgriffi...@gmail.com> >> wrote: >> >> >> >> Dear HCP peeps, >> >> >> I have a label file that I obtained from the BALSA database >> >> Gordon333_FreesurferSubcortical.32k_fs_LR.dlabel.nii >> >> ...and I would like to obtain the coordinates of the labels; and ideally >> convert it into a nifti volumetric image. >> >> I can't figure out what the correct wb_command calls are to do this. >> >> I assume this would need another file in addition, but I'm not sure which >> one is correct. >> >> The size of this label file is 91282, which I believe comprises ~30k >> vertices in left and right hemispheres, and ~30k subcortical voxels. But >> I'm not 100% sure for that. >> >> I know of gifti files in the HCP database with '32k' in the name that >> would seem to indicate the coordinates for the cortical surfaces. But I >> don't know about the additional 30k voxels (?). >> >> Simple problem; solution thus far escapes me. >> >> Could you please advise? >> >> >> >> >> Thanks, >> >> >> john >> >> >> >> >> >> >> >> >> >> >> >> >> >> >> -- >> >> Dr. John Griffiths >> >> Post-Doctoral Research Fellow >> >> Rotman Research Institute, Baycrest >> >> Toronto, Canada >> >> and >> >> Honorary Associate >> >> School of Physics >> >> University of Sydney >> >> _______________________________________________ >> HCP-Users mailing list >> HCP-Users@humanconnectome.org >> http://lists.humanconnectome.org/mailman/listinfo/hcp-users >> >> >> > > > -- > > Dr. John Griffiths > > Post-Doctoral Research Fellow > > Rotman Research Institute, Baycrest > > Toronto, Canada > > and > > Honorary Associate > > School of Physics > > University of Sydney > _______________________________________________ HCP-Users mailing list HCP-Users@humanconnectome.org http://lists.humanconnectome.org/mailman/listinfo/hcp-users
