Hello again,
I am wondering if one has behavioural regressors in an event-related design and you use 3dDeconvolve to find the BR coefficients and t-statistics, is it proper to transform each of these t-stats to the cortical surface using 3dVol2Surf and then continue with smoothing etc. or must you start with the +orig functional data? If this shortcut can be done do you know of any examples of this in the literature? I really appreciate your knowledge. -Linda
Hi Linda,
It is likely people do that, but the main point of
running a surface analysis is restricting it to the
gray matter, particularly including blurring and
clustering while respecting the topology.
Consider how afni_proc.py performs a surface
analysis via example 8 from the -help output.
Or add the ‘surf’ block and the 2 needed -surf_*
options to some other appropriate example.
- rick
Hi again,
I understand using the grey matter only reduces error and smooths better. However, I wonder if you must use afni_proc.py for a surface analysis or can you simply use ALL the regular afni commands like 3dcalc, 3dDeconvolve, FDR etc. except for the special surface commands like SurfSmooth etc. but what would you input into these commands-- will they assume a surface dataset is the same as any usual input? Also, by using 3dVol2Surf we convert timeseries datasets to what? a statistic on each surface node? If, that is true how do you do 3dDeconvolve after this command? I am sorry but I am just learning about this stuff and it seems harder than I thought—confusing I suppose. I really appreciate any help you can give.
Thanks everyone…
Hi experts,
Are my questions too obvious or too difficult. I don’t know where to find the answers directly but I could try things with suma now but could still use your knowledge and help. Appreciate you guys much.
-Linda
Hi Linda,
Catching up here, sorry, thanks for the reminder.
Using afni_proc.py is merely a convenience for
single subject analysis. It really just writes
a processing script that still uses a both
surface and volumetric programs to do the work.
Volumetric data gets mapped to the surface using
3dVol2Surf just after volumetric registration.
So time series data gets mapped to the surface.
Then blurring, scaling and regression can be
done on the surface. After all of that, beta
weights and t-stats are on the surface, and one
is ready for a group analysis.
Most AFNI programs that have no spatial concerns
(e.g. 3dcalc, 3dDeconvolve, 3dMVM, 3dFDR) should
not care whether data is on the surface or in
the volume. They worth the same.
Clustering and blurring are spatial computations,
so 3dmerge and 3dClustSim would not be appropriate
for surface data, for example.
Look at a surface analysis script to see what it
is doing. One example is s13.proc.FT.surf.
- rick
Thanks for your comments. They helped A LOT!!!
I have a question about program SurfClust. What I want to do is find the maximum node in a cluster in an ROI. I know you can draw ROIs in suma although I don’t know how at this moment. I was wondering that it seems you can output an ROI but not input one (unless I am wrong). We were thinking about inputing a parcellation from the human brainnetome atlas (which is brand new) and finding the largest node in this parcellation in hopes to get the corresponding coordinates in Volume space for each subject and averaging over subjects to get the “best” position in the brain region.
If you have any advice on the subject I would be very appreciative.
-Linda
Hello AFNI experts,
As I stated in the last message I am interested in using SurfClust. However I would like to use an atlas of parcellations. I am guessing other people would like to use their favorite atlas as well to define the clustering boundaries. I suppose you can convert the parcellation from Volume space to surface space and create a parcellation in cortical surface space but what would you use on the command line of SurfClust to obtain the clusters and target nodes within that parcellation? I am new to this stuff so I hope it is easier then I think it is. I appreciate your attention. -Linda
Hi Linda,
Maybe you just want to run SurfClust and get the max value
(among other things) for each cluster. For example, using
the class data under surf_demo/afni (note that surface #2
has correlation values):
SurfClust -spec …/SurfData/SUMA/std.DemoSubj_lh.spec
-surf_A smoothwm -input TS.lh.DEL.niml.dset 2 -thresh 0.6 -rmm -1
- rick
Hi Linda,
When using something like the FreeSurfer parcellation,
this is not a clustering step but an ROI step. It should
be enough to run 3dROIstats (3dmaskave, or whatever) on
such datasets. For example, from the same directory:
set hem = lh
3dVol2Surf -spec ../SurfData/SUMA/std.DemoSubj_${hem}.spec \
-surf_A std.${hem}.smoothwm.asc \
-surf_B std.${hem}.pial.asc \
-sv DemoSubj_SurfVol_Alnd_Exp+orig \
-grid_parent DemoSubj_EccExpavir.DEL+orig \
-map_func ave \
-f_steps 10 \
-f_index nodes \
-oob_value 0 \
-out_niml r.v2s.${hem}.DEL.niml.dset
3dROIstats -minmax -mask ../SurfData/SUMA/std.lh.aparc.a2005s.annot.niml.dset r.v2s.lh.DEL.niml.dset
How does that seem?
- rick
Thanks much Rick!
I know so little that I didn’t think you could do that probably because I thought there should be an -input ROI option in SurfClust if it were possible. sorry! Somehow I thought the surfaces were “special”. I also think the coordinates corresponding to the maximum node in the brain region can be found in AFNI or volume space and averaged over subjects to get the ‘best’ position for that brain region. This is my goal. thanks very much for your help. -Linda