Doing things with the nifti output of -cifti-convert is not what we mean by "volume-based analysis". What we mean by volume-based analysis is relying on volume-based registration to align the cortex, and combining same-index voxels across subjects in an attempt to get statistics or other results for a particular piece of cortex. This approach is undesirable, as volume-based registration does not correctly align cortex with normal variability of folding patterns, so the same post-registration voxel in different subjects can refer to different parts of cortex (or something entirely outside of cortex). It also is not well suited to aligning cortex based on features other than anatomical shape.
When you use -cifti-convert, you are still using grayordinates, the programs running on that data just don't know it. All -cifti-convert does is stuff the data matrix from the cifti file into the nifti-1 format so that tools meant for volume files can read it, no resampling or mapping or other conversion is done, so the values related to cortex still represent data on the surface, not in voxels. Thus, nothing you can do on the converted files is actually volume-based analysis, hence why it has been called "fake nifti". The downside to this is that you can't use neighborhood/spatial information while using a fake-nifti file, so you can't do smoothing, gradient, resampling, etc. The advantage is, since the cortex has been registered with surface-based registration, you can combine the same-index vertex (represented in a fake-nifti file by a "fake voxel", if you will) across subjects and get a valid result, as each vertex represents the same cortical location across subjects, and is never something outside of cortex. We generally only use one grayordinate space in the HCP released data, so any HCP cifti file that has a grayordinates mapping should be usable as the template, including the one you mentioned. Tim On Mon, May 11, 2015 at 12:12 PM, Carolina Ramirez < [email protected]> wrote: > Hi, > > I'm doing volume based analysis (with non-spatial relationships) of task > fMRIby converting the volumes in > MNINonlinear/Results/taskname/taskname_Atlas.dtseries.nii CIFTIs to NIFTI > using wb_command -cifti-convert. > > First, since we want to do group level analysis, I'm wondering if it's > correct or not to use the standard Conte69 > '91282_Greyordinates.dscalar.nii' CIFTI as a template in the conversion > back from CIFTI to NIFTI, given the comments in this thread: > https://www.mail-archive.com/hcp-users%40humanconnectome.org/msg00601.html > > Second, I understand that the purpose of creating the CIFTI format and the > higher level pipelines is to be able to do all the processing and analysis > in grayordinate space. Also, in a more recent thread you mention that some > people have had success with this "fake nifti" approach. > It's probably out of the scope of the HCP project, but still, can I ask > you if at some point you did some comparison between the results of a > 'cifti to nifti' volume based analysis with the grayordinates space > analysis? > In theory both approaches should yield somewhat equivalent results, but > the Pipelines are optimized for grayordinates, so it would be reasonable to > expect that the HCP task analysis Pipelines will give better(more > complete?) results in general, right? > > Thanks, > > Carolina. > > > _______________________________________________ > 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
