censoring

AFNI Message Board
1

Hi!

I have a subject with quite a lot of censoring. However, the following afni_proc works well:

afni_proc.py -subj_id sub11 -script proc.sub11 -scr_overwrite -blocks
despike tshift align tlrc volreg blur mask regress -copy_anat
/anatomy/sub11.nii -dsets
/resting_state/sub11_1.nii
/resting_state/sub11_2.nii
-tcat_remove_first_trs 0 -align_opts_aea -giant_move -volreg_align_to
MIN_OUTLIER -volreg_align_e2a -volreg_tlrc_warp -blur_size 4.0
-regress_censor_motion 0.25 -regress_bandpass 0.01 0.1
-regress_apply_mot_types demean deriv -regress_motion_per_run
-regress_est_blur_errts -regress_run_clustsim no

Now, I’m trying to understand why it crashes when I try a similar thing, but also regressing out signal from white matter and vent. I get the message of: ** ERROR: total number of fixed regressors (250) is too many for 249 retained time points!
Is there a way to save this subject regressing out wm and ventricles? Here is the afni_proc code that leads to the error:

afni_proc.py -subj_id sub11 -script proc.sub11 -scr_overwrite -blocks despike
tshift align tlrc volreg blur mask regress -copy_anat
/SUMA/sub11/T1.nii.gz -dsets
/resting_state/sub11_1.nii
/resting_state/sub11_2.nii
-anat_follower_ROI aaseg anat
/SUMA/sub11/aparc.a2009s+aseg.nii.gz
-anat_follower_ROI aeseg epi
/SUMA/sub11/aparc.a2009s+aseg.nii.gz
-anat_follower_ROI FSvent epi
/SUMA/sub11/fs_ap_latvent.nii.gz
-anat_follower_ROI FSWe epi
/SUMA/sub11/fs_ap_wm.nii.gz
-anat_follower_erode FSvent FSWe -tcat_remove_first_trs 0
-align_opts_aea -giant_move -volreg_align_to MIN_OUTLIER
-volreg_align_e2a -volreg_tlrc_warp -blur_size 4.0
-regress_censor_motion 0.25 -regress_bandpass 0.01 0.1
-regress_apply_mot_types demean deriv -regress_motion_per_run
-regress_est_blur_errts -regress_run_clustsim no -regress_ROI_PC FSvent
3 -regress_ROI_PC_per_run FSvent -regress_make_corr_vols aeseg FSvent
-regress_anaticor_fast -regress_anaticor_label FSWe

I appreciate any input. Thanks!

2

Hola Ana,

The error indicates that the number of nuisance regressors is larger than the number of time points after censoring. In other words, the remaining data would have -1 degrees of freedom!!

This is probably due to the combination of censoring and bandpass filtering. What is the TR of the acquisition? If the TR is too fast, bandpass filtering needs to include many, many regressors to remove all the frequencies from 0.1 to the 1/(2*TR) (i.e. the nyquist frequency).
Alternatively, you might also try to increase the censoring threshold to 0.3 and see if this helps.

Best wishes,
Cesar

3

Thanks, Cesar! Changing the censoring threshold to 0.3 helped, as you suggested. I’ve decided to exclude this subject at the end. Thanks a lot for your message anyway because this is clear to me now.