How to run 3dVol2Surf after progressing by afni_proc.py

AFNI Message Board
1

Hello，
I want use this below .sh to project the volumetric data to the surface for better group analysis statistics.
All of my subjects have completed preprocessing (-blocks tshift align tlrc volreg blur mask scale regress）by afni_proc.py and have been registered to MNI space. I want to use 3dVol2Surf to do group analysis statistics on the surface space.

foreach subj (sub01 sub02 .....)
foreach hemi ( lh rh )
  3dVol2Surf \
      -spec ${surfData}/std.141.MNI152_2009_${hemi}.spec \
      -surf_A smoothwm \
      -surf_B pial \
      -sv ${surfData}/MNI152_2009_SurfVol.nii \
      -grid_parent ${subData}/stats.${subj}+tlrc \
      -map_func ave \
      -f_steps 10 \
      -f_index nodes \
      -out_niml ${outData}/std.141.${hemi}.${subj}.niml.dset
end
end

my questiond are these:
1)I’ve seen some other examples of 3dVol2Surf where the anatomical and functional runs appear to still be in native space.However,my data have been aligned to to MNI space,I just want to know if it could be accepted by these step that "registering to MNI space in volume(I have done)"->"run 3dVol2Surf"->''calcu group analysis statistics"?

2)If these step in question1) could be accept,(my standard space anatomical is mni152_2009), is it right to use this code?

2

Howdy-

My two cents worth here, and others will hopefully chime in:

Q1) The "other examples of 3dVol2Surf where the anatomical and functional runs appear to still be in native space" probably refers to including projection to the surface within afni_proc.py (AP) itself, which is possible by inserting the "surf" block (and appropriate suboptions like pointing at the white matter surfaces, setting your blur radius for surface extent, etc.), and removing the "tlrc" and "mask" blocks (and associated suboptions). You can actually copy your AP command and leave virtually everything else unchanged, just doing the above.

  • IMHO, it is OK to do what you are doing---process all the way to a standard volume, and then project results at the end using standard surfaces that AFNI makes.

  • However, it would probably be better to include the surf block in processing directly, for a couple reasons:

    • Every subject's brain is a different, including different numbers of sulci and gyri in some locations. Performing projection to the surface from the subject's own anatomical volume to the surface mesh created on that subject's own anatomical should have higher fidelity of structural matching than projecting from the warped subject anatomical to a MNI-generated mesh.

    • When you include surface projection within AP itself, the blurring to the EPI data will be applied on the surface, rather than volumetrically; this means that the blurring should really be constrained to local GM only, rather than to include other tissue types and/or other, topologically distant (but geometrically close) gyral features. This is perhaps the main reason why surface-based analyses are popular.

So, all processing includes imperfect alignment and blurring of information, but when individual subject anatomical meshes are reasonably-well constructed, the project-to-surface-while-processing approach should have a couple advantages for cortical studies, at least in theory. (The downsides tend to be that not all regions of interest will be on an estimated mesh, like subcortical nuclei, and in practice the meshes aren't always perfect; they also require projection of information through the GM sheet, though this process is often reasonably straightforward.)

Q2) I'm guessing that you might actually know all of what I wrote in Q1, because your 3dVol2Surf command looks like it came directly from what AP would generate if you included the 'surf' block within the command itself. That seems reasonable to me, yep.

One subtle consideration might be what operation to perform for -map_func .. since you are mapping the stats. I do think 'ave' is what I would probably pick still, but I would consider 'max_abs', as well, to get the peak statistic value from within the GM (I don't think the projected value itself will be absolute-valued, and you should verify that). However, in general, the voxel sizes are pretty large compared to cortical thickness, so this might not matter too much, anyways, though.

--pt

4

Thanks,ptaylor

I know what you said "Every subject's brain is a different, including different numbers of sulci and gyri in some locations. Performing projection to the surface from the subject's own anatomical volume to the surface mesh created on that subject's own anatomical should have higher fidelity of structural matching than projecting from the warped subject anatomical to a MNI-generated mesh", actually，that's I am concerned ，too.
I have done this "registering to MNI space in volume(I have done)"->"run 3dVol2Surf"->''calcu group analysis statistics" this afternoon ，however，the result was not satisfied，I was a little sad. (I would also consider your suggestion about -map_func.) I guess may be it had lost of some validated data by projecting from the warped subject anatomical to a MNI-generated mesh. I'm going to re-do the data preprocessing with afni_proc.py including projection to the surface possibly tomorrow.

I have another question that we can do multiple comparison correction via clustSim in 3dTtest, but is there a clustSim-like method for correction in the surface space other than FDR?

thanks!
-fuying

5

Hi, Fujing-

It sounds like it does make sense for you to run AP with the 'surf' block. I'll note some of the online AFNI Academy Bootcamp videos about surface visualization in SUMA, and here is Rick teaching at a Bootcamp we had at MIT, about afni_proc.py and SUMA specifically (videos 26-28), with 24-25 covering general SUMA).

For cluster estimation on the surface, there is this program:
slow_surf_clustsim.py

--pt

6

Thanks for your help, ptaylor. I will try that.
-fuying

7

Hello，pt
I redid the preprocessing for adding surf and the code in AP is as follows.However, I found some problems, some of my subjects have very low TSNR of 50, but the same subjects in the volume space based preprocessing results report TSNR is not so low (about 140).
There are no errors reported in the whole preprocessing runtime, so I want to know

  1. why there are such results conduct different TSNR?
    2)Others subjects' TSNR are around 120-160,is it accept?
 afni_proc.py -subj_id sub08 -script proc.Visual_surf_pro -scr_overwrite                                                          \
     -blocks tshift align volreg surf blur scale regress -copy_anat                                                               \
     /media/newVisual_Allsub_anat/sub08_T1.nii                                              \
     -dsets                                                                                                                       \
     /media/newVisual_Allsub_EPI/sub08_newVisual_run1+orig.HEAD                             \
     /media/newVisual_Allsub_EPI/sub08_newVisual_run2+orig.HEAD                             \
     /media/newVisual_Allsub_EPI/sub08_newVisual_run3+orig.HEAD                             \
     /media/newVisual_Allsub_EPI/sub08_newVisual_run4+orig.HEAD                             \
     /media/newVisual_Allsub_EPI/sub08_newVisual_run5+orig.HEAD                             \
     /media/newVisual_Allsub_EPI/sub08_newVisual_run6+orig.HEAD                             \
     /media/newVisual_Allsub_EPI/sub08_newVisual_run7+orig.HEAD                             \
     /media/newVisual_Allsub_EPI/sub08_newVisual_run8+orig.HEAD                             \
     -surf_anat                                                                                                                 \
     /media/V2_Vol2Surf/SUBJ_sub08/SUMA/SUBJ_sub08_SurfVol.nii     \
     -surf_spec                                                                                                                   \
     /media/V2_Vol2Surf/SUBJ_sub08/SUMA/std.141.SUBJ_sub08_lh.spec \
     /media/V2_Vol2Surf/SUBJ_sub08/SUMA/std.141.SUBJ_sub08_rh.spec \
     -tcat_remove_first_trs 0 -align_opts_aea -cost lpc+zz -giant_move                                                            \
     -check_flip -volreg_align_to MIN_OUTLIER -volreg_align_e2a -blur_size                                                        \
     4.0 -regress_stim_times                                                                                                      \
     /media/stimuliOrder/stim_Exp1newVisual_far.txt                                         \
     /media/stimuliOrder/stim_Exp1newVisual_near.txt                                        \
     /media/stimuliOrder/stim_Exp1newVisual_Sfar.txt                                        \
     /media/stimuliOrder/stim_Exp1newVisual_Snear.txt                                       \
     -regress_stim_labels far near Sfar Snear -regress_basis 'BLOCK(16,1)'                                                        \
     -regress_censor_motion 0.3 -regress_motion_per_run -regress_opts_3dD                                                         \
     -gltsym 'SYM: near - far' -glt_label 1 near-far -regress_compute_fitts                                                       \
     -regress_make_ideal_sum sum_ideal.1D

My out.ss_review.sub20.txt is as follows.

subject ID : sub20
AFNI version : AFNI_22.0.01
AFNI package : linux_ubuntu_16_64
TR : 2.0
TRs removed (per run) : 0
num stim classes provided : 4
final anatomy dset : anat_final.sub20+orig.HEAD
final stats dset : stats.sub20.rh.niml.dset
orig voxel counts : 104 104 64
orig voxel resolution : 2.000000 2.000000 2.000000
orig volume center : 1.978203 -10.691002 4.377831
final voxel resolution : 1.000000 1.000000 1.000000

motion limit : 0.3
++ WARNING: file /home/linux/.afni.log is now 276005427 (276 million) bytes long!
num TRs above mot limit : 1
average motion (per TR) : 0.026544
++ WARNING: file /home/linux/.afni.log is now 276005584 (276 million) bytes long!
average censored motion : 0.0262392
max motion displacement : 4.95706
max censored displacement : 4.91663
average outlier frac (TR) : 0.0197054
flip guess : NO_FLIP

num runs found : 8
num TRs per run : 161 161 161 161 161 161 161 161
num TRs per run (applied) : 161 161 161 161 161 161 161 159
num TRs per run (censored): 0 0 0 0 0 0 0 2
fraction censored per run : 0 0 0 0 0 0 0 0.0124224
TRs total (uncensored) : 1288
TRs total : 1286
degrees of freedom used : 84
degrees of freedom left : 1202

TRs censored : 2
censor fraction : 0.001553
num regs of interest : 4
num TRs per stim (orig) : 346 349 354 355
num TRs censored per stim : 2 0 2 0
fraction TRs censored : 0.006 0.000 0.006 0.000

ave mot per sresp (orig) : 0.028746 0.023563 0.024607 0.024814
ave mot per sresp (cens) : 0.027643 0.023563 0.023506 0.024814

TSNR average : 50.8348
++ WARNING: file /home/linux/.afni.log is now 276008715 (276 million) bytes long!
maximum F-stat : 99.5006
This is the code to calculate TSNR in AP：

 foreach hemi ( lh rh )
    3dTstat -mean -prefix rm.signal.all.$hemi.niml.dset                       \
        all_runs.$subj.$hemi.niml.dset"[$ktrs]"
    3dTstat -stdev -prefix rm.noise.all.$hemi.niml.dset                       \
        errts.${subj}.$hemi.niml.dset"[$ktrs]"
    3dcalc -a rm.signal.all.$hemi.niml.dset                                   \
           -b rm.noise.all.$hemi.niml.dset                                    \
           -expr 'a/b' -prefix TSNR.$subj.$hemi.niml.dset
end
8

Hello,

One difference between volume and surface analysis is the blurring method and its corresponding FWHM. In the volume case, -blur_size is an added Gaussian blur. In the surface case, -blur_size is the FINAL estimated FWHM, it is a blur to fwhm parameter, rather than a blur added one.

Anyway, so the surface analysis is less blurry here, which might contribute to lower TSNR. Also, TSNR is often lower along the cortical ribbon than in white matter, for example, possible due in part to motion.

For an alternate comparison, you might add -volreg_compute_tsnr, which would not include any blurring (or regression, two big differences). At least that would give something for comparison, even if it is not a great one.

  • rick