issues about TRs that excluded due to motion censoring and BOLD signals extracted by all run dataset

AFNI Message Board
1

Hello, everyone!
For ROI analysis, I extracted the BOLD signals from the all_runs dataset that created after the preprocessing steps with afni_proc.py.
Now I wonder whether the signals of TRs, that is excluded due to motion censoring, remain the same without any processing (such as, interpolation) in the all_runs dataset.
So what should I do if I want to process (interpolation) the signals of TRs, that is excluded in the GLM analysis, in the all_runs dataset?

My preprocessing script was like:

 afni_proc.py -subj_id ${subj} -script proc. -scr_overwrite -blocks tshift                                          \
     align tlrc volreg blur mask scale regress -copy_anat                                                            \
     &datadir/Mean_anat_sub02_TPL.nii   \
     -dsets                                                                                                          \
     &datadir/sub02_run1+orig.HEAD  \     
    &datadir/sub02_run2+orig.HEAD  \
     &datadir/sub02_run3+orig.HEAD \
     -tcat_remove_first_trs 0 -align_opts_aea -giant_move -tlrc_base                                                 \
    MNI152_T1_2009c+tlrc -volreg_align_to MIN_OUTLIER -volreg_align_e2a                                             \
     -volreg_tlrc_warp -blur_size 4.0 -regress_stim_times                                                            \
     &datadir/sub02_1.txt \
    &datadir/sub02_2.txt \
     &datadir/sub02_3.txt \
     &datadir/sub02_4.txt \
    -regress_stim_labels ET ST ENT SNT -regress_basis 'BLOCK(8,1)'                                                  \     
 -regress_censor_motion 0.3 -regress_motion_per_run -regress_opts_3dD                                            \
     -gltsym 'SYM: ET +ST -ENT -SNT' -glt_label 1 T
     -gltsym 'SYM: ET +ENT -ST -SNT' -glt_label 2 S
     -regress_compute_fitts                                                             \
     -regress_make_ideal_sum sum_ideal.1D -regress_est_blur_epits                                                    \
    -regress_est_blur_errts

Thank you!
Zhiqing

2

Hello,

Before I forget, if this is for an ROI analysis, then you probably do not want a 'blur' block included. Removing it would avoid polluting the ROIs.

Also, this uses just an affine registration to the template. Using non-linear registration via @SSwarper would likely provide much better anatomy to atlas correspondence for any ROI averaging. The output transformation from @SSwarper could be passed to afni_proc.py to apply to the EPI data.

But just to understand, exactly what will be done to perform the ROI analysis with the all_runs data? In the afni_proc.py analysis, you have included stim information for a task-based voxelwise analysis. Will you extract ROI averages from the all_runs time series to then run similar linear regression on? It is not clear how the task would be applied here.

Thanks,

  • rick
3

Oh, and I forgot to actually answer your question.

The all_runs time series is not censored at all. Censoring is done in the linear regression step (with all_runs as input, say). The censored time points would be all zero in the resulting errts (error or residual time series) output from the regression.

  • rick
4

Hi, Rick.
Thank you so much for your early reply!
After preprocessing, ROI analysis included extracting the BOLD signals around the peak value and performing ANOVAs within each ROI. Also I have done preprocessing steps without blur and next I will extract the BOLD signals from all runs dataset and apply MVPA.
I just set the BOLD signals of TRs that should be censored as nan and then averaged blocks of same condition, ignoring the nan values, in ANOVA analysis. But for MVPA, I wonder if I should do a interpolation step for the censored TRs with 3dTproject command.

Zhiqing

5

Hi Zhiqing,

I am not sure what MVPA software you are using, but it might be good to ask the developers what the expectations for the input time series are. The software might be making assumptions that you are not aware of. Interpolation is probably not a recommended step, but that would make the choice between leaving in zero volumes (which is probably not preferable), removing them (if time discontinuities are okay) or not censoring at all.

But keep in mind that if you are working with the all_runs data, then no censoring or even regression has been applied to it.

  • rick
6

thank you for your answer！
I did not use any software for MVPA for BOLD signals (that is what we are intereasted), just wrote the script myself. I decide not to censor any TRs, because there are not much TRs that should be censored.
Thank you again！