The geodesic distances are something we compute on the fly during processing, they aren't in a file (we don't need to know the distance to things that are really far, so it is a waste to compute or store all of them). You can get the distances from one vertex at a time with wb_command -surface-geodesic-distance for individual subjects, other than that we don't currently have a good way to export them for other software.
They can be computed from the information in surface files, but you basically need to use a language that is fast for loops (if you want it in matlab, you'll probably want to look at .mex files). We use dijkstra for this, with data structures suitable for the rather sparse graphs that represent surfaces (32k vertices, but only 6 edges from a vertex for the naive geodesic distances). Tim On Tue, Jun 6, 2017 at 5:58 PM, Manasij Venkatesh <[email protected]> wrote: > Thank you for your quick response, that was really helpful. I'll use the > 180 parcellations as well, but I wish to do some analysis using custom ROIs. > > A quick follow up question: Suppose I want to do clustering using these > geodesic distances, is there a file containing them as well? > > Sincerely, > Manasij > > > On Tue, Jun 6, 2017 at 6:12 PM, Timothy Coalson <[email protected]> wrote: > >> First, you might consider using our 180-area-per-hemisphere functional >> parcellation (https://balsa.wustl.edu/study/show/RVVG), which is more >> specialized to human cortex than a distance-based clustering would be. >> Also keep in mind that vertices being close in 3D doesn't mean they are >> related, they can be on the opposite side of a large sulcus - we use >> geodesic distance for this purpose (crawling along the surface triangles). >> Additional answers inline. >> >> Tim >> >> On Tue, Jun 6, 2017 at 4:55 PM, Manasij Venkatesh <[email protected]> >> wrote: >> >>> Hi, >>> >>> The number of vertices on the surface is of the order of ~60k (excluding >>> subcortical regions). I wish to work in an ROI space (say 100 ROIs) by >>> clustering the cortical vertices that are near-by. To do so, I need to know >>> the co-ordinates of the vertices on the surface. I had the following >>> questions regarding this: >>> >>> 1. Does the data in *.dtseries.nii correspond to the time series of the >>> vertices after surface alignment? If so, which is the file containing the >>> coordinates of these vertices? >>> >> >> Yes, the coordinates are in ".surf.gii" files. You probably want >> midthickness coordinates, but there are also pial and white matter >> surfaces. The MNI space versions are in MNINonLinear/fsaverage_LR32k, but >> there are also structural space versions (no scaling or stretching from >> distortion corrected scan data) in the T1w folder. >> >> >>> 2. Do the vertices across patients correspond to the same region in the >>> brain? >>> >> >> That is the purpose of registration, so fs_LR surfaces have this >> property, yes. The correspondence is better with MSMAll than with MSMSulc >> or FreeSurfer. >> >> >>> If this is true, shouldn't there be one surface file for all subjects >>> after alignment? I'm wondering why there's a surface file for each subject. >>> >> >> No, surface registration does not do anything to anatomical shape. After >> registration and resampling of a subject's surface, it looks exactly the >> same as before, it is simply that when we click on a point on one subject's >> surface, we can now find a matching point on another subject's surface. >> None of this requires altering the anatomical shape of any subject, so they >> all still have individual surfaces. >> >> One important reason for this is that many functional areas are not >> strictly matched to nearby sulci/gyri (also, different subjects have rather >> different folding patterns in a fair portion of cortex). Thus, MSMAll >> establishes correspondence based on areal features (myelination, visual >> areas, rfMRI components, cortical thickness), which causes folding patterns >> to not have as good correspondence across subjects as they would with >> MSMSulc, which establishes correspondence based only on folding patterns. >> >> >>> Thank you in advance. >>> >>> Sincerely, >>> Manasij >>> >>> _______________________________________________ >>> HCP-Users mailing list >>> [email protected] >>> http://lists.humanconnectome.org/mailman/listinfo/hcp-users >>> >> >> > _______________________________________________ HCP-Users mailing list [email protected] http://lists.humanconnectome.org/mailman/listinfo/hcp-users
