As has been discussed on the list before, group average surfaces do not have the geometric definition that volume spaces expect:
https://www.mail-archive.com/hcp-users%40humanconnectome.org/msg02669.html The AAL atlas appears to have been defined from a single individual - if this is correct, then you should use the surfaces from that individual (if available) to map it from volume to surface. You can then use surface registration to apply it to other subjects or group average surface data. If the surface for that individual is not available, then you could try using a set of individual MNI space surfaces, and combining the results. Using any group average surface for volume to surface mapping is basically a last resort, but you may have better luck with one derived from folding pattern alignment (MSMSulc or freesurfer registered), while I believe the latest HCP group average data uses areal feature registration, and so our group average surface has even less geometric definition. Tim On Mon, Jul 25, 2016 at 3:02 AM, Salim Arslan <[email protected]> wrote: > Hi, > > We are evaluating brain parcellation schemes (anatomical atlases, > connectivity-driven parcellations, functional segmentations etc) using > data from the HCP. Some parcellations are originally volume-based, hence > need to be sampled to surface. > > My question is, what is a good way to project the AAL atlas from the > volume onto the 32k average surface? Besides many surface based atlases, > such as the ones provided by the HCP, we specifically wish to include the > AAL to our experiments, as it is a popular atlas used in network analysis. > > What we have done so far is that we projected the AAL labels to a > 32K_fs_LR surface using the FreeSurfer mri_vol2surf function in the > following setting: > > *mri_vol2surf --mov $AAL --mni152reg --hemi lh --surf $SURF --o > $SURFACE_NAME.asc.w --out_type paint --trgsubject fsaverage* > > where $AAL corresponds to ROI_MNI_V5.nii, the last version of the ALL > atlas, provided by http://www.gin.cnrs.fr/AAL and $SURF corresponds to a > group average mid. thickness surface, provided by the HCP as part of the > tutorial dataset (HCP_Q1_Q6_GroupAvg_Related440_Unrelated100_v1). > > After converting the output surface file in FreeSurfer format to gifti > (shape.gii), the final cortical parcellation was not perfect but still > seemed to be usable. We further applied some post-processing to each parcel > (erosion/dilation) to smooth the boundaries and managed to obtain 83 > parcels out of 90 of the original atlas. I am attaching screenshots of the > original and post-processed cortical parcellations after sampling (left > hemisphere). > > The thing you may have also noticed is that some of the cortical > parcellation boundaries do not follow the sulci course in the cortical > surface. How would the experts in the mailing list interpret this? Is it > kind of a side effect we should expect after such a projection from volume > to surface? Can you suggest (if there is) a more elegant way of doing this > sampling? > > Any feedback is greatly appreciated. Thanks for your attention. > > Kind regards, > Salim > > _______________________________________________ > 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
