As people working with HCP CIFTI images know, the cortex is represented by (surface) vertices, but subcortical structures as voxels. As I explained before, you can use the wb_command -cifti-separate to make a volumetric NIfTI of a subcortical structure (say AMYGDALA_LEFT). The HCP minimally preprocessing pipelines create the CIFTI versions of each (task, in my case) run and also volumetric NIfTI versions. So my question: is timecourse of a voxel in a subcortical structure the same when taken out of an unpacked CIFTI and a NIfTI? Answer: no.
AMYGDALA_LEFT via wb_command -cifti-separate (green stripe), and the _AtlasSubcortical_s2.nii.gz image (yellow stripe) for one of our HCP pipeline minimally-processed task fMRI runs.
The _AtlasSubcortical_s2.nii.gz image was created by the pipeline, and looks to be the subcortical (volumetric) part of the CIFTI: the value at this voxel is identical (green and yellow). The value for the voxel in the volumetric NIfTI (red) is a bit different, however.
Note that I don't think this mismatch is an error: parcel-constrained smoothing is done during the HCP pipelines that make the CIFTIs (e.g., Glasser et al. 2013), and I suspect that this smoothing accounts for the different voxel values. However, it's good to be aware of the additional processing: if you're doing a volumetric analysis with HCP-type preprocessing, it matters whether you take the subcortical structures out of the CIFTI or the NIfTI.