some question about TENT regress in the analysis of event design

AFNI Message Board
1

Hello, any experts,

I have some questions in regress_reml_exec. My experiment is event design, I try to use the following AP to process.

Blockquote
afni_proc.py -subj_id sub001 -script
proc.step1_Grouping_processing -scr_overwrite -blocks tshift align tlrc volreg blur mask scale regress
-copy_anat
path/anat/sub001_SSwarper/anatSS.sub001.nii
-dsets
path/EPI/sub001_task_run1+orig.HEAD
path/EPI/sub001_task_run2+orig.HEAD
path/EPI/sub001_task_run3+orig.HEAD
path/EPI/sub001_task_run4+orig.HEAD
-tcat_remove_first_trs 0 -align_opts_aea -giant_move -tlrc_base \
MNI152_2009_template_SSW.nii.gz -tlrc_NL_warp -tlrc_NL_warped_dsets \
path/anat/sub001_SSwarper/anatQQ.sub001.nii
path/anat/sub001_SSwarper/anatQQ.sub001.aff12.1D
path/anat/sub001_SSwarper/anatQQ.sub001_WARP.nii
-volreg_align_to MIN_OUTLIER -volreg_align_e2a -volreg_tlrc_warp \
-blur_size 6.0 -regress_stim_times
patha/sub001/cona.txt
patha/sub001/conb.txt
patha/sub001/Nocona.txt
patha/sub001/Noconb.txt
-regress_stim_labels cona conb Nocona Noconb
-regress_basis 'TENT(0,14,15)'
-regress_anaticor -regress_reml_exec -regress_censor_motion 0.3 - regress_motion_per_run \
-regress_opts_3dD -gltsym 'SYM: cona -conb' -glt_label 1 cona-conb
-gltsym 'SYM: Nocona -Noconb' -glt_label 2 Nocona-Noconb \
-regress_compute_fitts -regress_make_ideal_sum sum_ideal.1D \
-regress_est_blur_epits -regress_est_blur_errts

I have some question about the AP：
1）-regress_anaticor -regress_reml_exec and other parameters is acceptable to be used in the event design （TENT（0,14,15）TR=1s，voxel =2.5 2.52.5 and per trail is 9s）？
2）if the question1 is ok;I have just run 2 subjects using this AP,and I have get the stats.sub+tlrc. stats.sub_REML+tlrc. and stats.sub_REMLvar+tlrc. ,so shoud I use the beta value from the stats.sub_REML+tlrc if I want to do any further analysis about beta value ?
3)if stimu_IM is also accessible in the TENT if I want to caculate voxelwise value for mutiplevalue pattern analysis(MVPA) ? or there is someother parameter could be used just like stimu_IM?

2

I will let @Gang reply, but I just wanted to make a slightly more vertically spaced+aligned version of the AP command, to help read when addressing the Qs:

afni_proc.py                                               \
    -subj_id                 sub001                        \
    -script                  proc.step1_Grouping_processing \
    -scr_overwrite                                         \
    -blocks                  tshift align tlrc volreg blur \
                             mask scale regress            \
    -copy_anat               path/anat/sub001_SSwarper/anatSS.sub001.nii \
    -dsets                   path/EPI/sub001_task_run1+orig.HEAD \
                             path/EPI/sub001_task_run2+orig.HEAD \
                             path/EPI/sub001_task_run3+orig.HEAD \
                             path/EPI/sub001_task_run4+orig.HEAD \
    -tcat_remove_first_trs   0                             \
    -align_opts_aea          -giant_move                   \
    -tlrc_base               MNI152_2009_template_SSW.nii.gz \
    -tlrc_NL_warp                                          \
    -tlrc_NL_warped_dsets    path/anat/sub001_SSwarper/anatQQ.sub001.nii \
                             path/anat/sub001_SSwarper/anatQQ.sub001.aff12.1D \
                             path/anat/sub001_SSwarper/anatQQ.sub001_WARP.nii \
    -volreg_align_to         MIN_OUTLIER                   \
    -volreg_align_e2a                                      \
    -volreg_tlrc_warp                                      \
    -blur_size               6.0                           \
    -regress_stim_times      patha/sub001/cona.txt         \
                             patha/sub001/conb.txt         \
                             patha/sub001/Nocona.txt       \
                             patha/sub001/Noconb.txt       \
    -regress_stim_labels     cona conb Nocona Noconb       \
    -regress_basis           'TENT(0,14,15)'               \
    -regress_anaticor                                      \
    -regress_reml_exec                                     \
    -regress_censor_motion   0.3                           \
    -regress_motion_per_run                                \
    -regress_opts_3dD        -gltsym 'SYM: cona -conb'     \
                             -glt_label 1 cona-conb        \
                             -gltsym 'SYM: Nocona -Noconb' \
                             -glt_label 2 Nocona-Noconb    \
    -regress_compute_fitts                                 \
    -regress_make_ideal_sum  sum_ideal.1D                  \
    -regress_est_blur_epits                                \
    -regress_est_blur_errts

Also, on a minor note when doing this, I noticed there was an erroneous space inserted between the - and regress_motion_per_run option name in the original posting. Just so you are aware, that that doesn't sneak into your actual runtime code.

--pt

3

Yeap， I'm glad to hear Gang's recommendations thank you ptaylo.
fuy

4

1）-regress_anaticor -regress_reml_exec and other parameters is acceptable to be used in the event design （TENT（0,14,15）TR=1s，voxel =2.5 2.5 2.5 and per trail is 9s）？

I suggest using TENTzero(0,14,15) unless you specifically aim to capture the BOLD response at the stimulus onset.

2）if the question1 is ok;I have just run 2 subjects using this AP,and I have get the stats.sub+tlrc. stats.sub_REML+tlrc. and stats.sub_REMLvar+tlrc. ,so shoud I use the beta value from the stats.sub_REML+tlrc if I want to do any further analysis about beta value ?

Yes, that sounds reasonable under most scenarios.

3)if stimu_IM is also accessible in the TENT if I want to caculate voxelwise value for mutiplevalue pattern analysis(MVPA) ? or there is someother parameter could be used just like stimu_IM?

If you want to capture BOLD response at the trial level, the tent basis function is likely not practically feasible. You may have to use GAM in 3dDeconvolve or GLMsingle.

Gang Chen

6

Thank you for your detailed answer, Gang, your answers help me greatly. I want to know why you say " If you want to capture BOLD response trial level, the tent basis function is likely not practically feasible. You may have to use GAM in 3dDeconvolve or GLMsingle."
By the way, I have done the gPPI for block design before, however, I don't know how to do the gPPI for the event design (just like the one above TENT(0,14,15)), Are there any specific differences in the steps of gPPI between the two designs? could you give me some advice?
thank you
fuying

7

Fuying,

I want to know why you say " If you want to capture BOLD response at the trial level, the tent basis function is likely not practically feasible. You may have to use GAM in 3dDeconvolve or GLMsingle."

Estimating the BOLD response at the trial level using the deconvolution approach presents two major challenges. First, the estimation is often unreliable due to the difficulty in disentangling the BOLD signal from mediation effects and various sources of noise. Second, numerical issues, such as collinearity, can further complicate the analysis.

I have done the gPPI for block design before, however, I don't know how to do the gPPI for the event design (just like the one above TENT(0,14,15)), Are there any specific differences in the steps of gPPI between the two designs?

The PPI approach consists of two critical steps: deconvolution and the creation of an interaction term. The deconvolution process depends on selecting an appropriate a priori hemodynamic response function (HRF), which is crucial for accuracy. Additionally, generating condition-level effects and interaction terms using the basis function approach introduces further challenges. While it's possible to use the estimated HRFs for both steps, I don't have experience with how this approach performs in practice.

Gang Chen