I thought someone else might answer this, but I'll give it a shot.

For the volume part of the data, there is some smoothing that happens (2mm
FWHM), originally as part of making the subcortical ROIs of each subject
conform to the cifti space.  The recent versions of the pipelines use a
different method now, but it still is set to do similar smoothing by
default, to reduce differences from the released HCP data.  This smoothing
occurs separately within each subcortical parcel, and the effects of having
an ROI edge near a voxel may result in an amplification of the smoothing
kernel.

For surface data, I know of two things that are going on.  First, when data
is mapped from volume to surface, it uses a weighted average between
several voxels, depending on how the ribbon intersects them, which results
in a reduction of noise.  Second, the surface data was smoothed by the same
distance as the subcortical data (2mm FWHM).

The surface FWHM calculation is also fairly naive, in that it assumes all
vertices have the same spacing, which might cause a bias in it.  We don't
use this measure much, if at all, so we haven't come up with a method that
takes this into account.

It may also be the case that the non-noise part of the smoothed fMRI signal
has less within-neighborhood variation in it than a similarly smoothed pure
noise image would have, which would also cause the FWHM estimation to be
larger than the combination of "smoothness" amounts would suggest.

Tim


On Wed, May 31, 2017 at 11:34 PM, Ely, Benjamin <[email protected]>
wrote:

> Hi everyone,
>
> I noticed recently that the smoothness of the minimally-preprocessed
> grayordinate-space HCP data is surprisingly high, especially at 3T. For
> subject 100610, as an example, wb_command -cifti-estimate-fwhm reports:
>
> rfMRI_REST1_LR_Atlas_MSMAll_hp2000_clean.dtseries.nii
> CORTEX_LEFT FWHM: 18.3385
> CORTEX_RIGHT FWHM: 16.5479
> Voxels FWHM: 7.3608, 7.1032, 6.95473
>
> By contrast, fsl smoothest gives the smoothness of the corresponding
> volumetric fMRI file as:
>
> rfMRI_REST1_LR_hp2000_clean.nii.gz
> FWHMx = 1.98571 mm, FWHMy = 1.84018 mm, FWHMz = 1.99221 mm
>
> That seems like a huge jump in smoothness, especially for the subcortical
> data, which is represented in essentially the same voxel space.
>
> The difference is smaller for the 7T data, but still pronounced:
>
> rfMRI_REST1_7T_PA_Atlas_1.6mm_MSMAll_hp2000_clean.dtseries.nii
> CORTEX_LEFT FWHM: 7.04818
> CORTEX_RIGHT FWHM: 6.99358
> Voxels FWHM: 4.63357, 4.54396, 4.23441
>
> rfMRI_REST1_7T_PA_hp2000_clean.nii.gz
> FWHMx = 1.7863 mm, FWHMy = 1.74615 mm, FWHMz = 1.66513 mm
>
> Do these results seem correct and if so, why are the grayordinate-space
> data (including the subcortical voxels) so much smoother?
>
> Many thanks,
> -Ely
>
> Example of full workbench command, from subject directory:
> wb_command -cifti-estimate-fwhm \
> MNINonLinear/Results/rfMRI_REST1_LR/rfMRI_REST1_LR_Atlas_
> MSMAll_hp2000_clean.dtseries.nii \
> -merged-volume -column 1 \
> -surface CORTEX_LEFT MNINonLinear/fsaverage_LR32k/
> 100610.L.midthickness_MSMAll.32k_fs_LR.surf.gii \
> -surface CORTEX_RIGHT MNINonLinear/fsaverage_LR32k/
> 100610.R.midthickness_MSMAll.32k_fs_LR.surf.gii
>
> _______________________________________________
> 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

Reply via email to