afni.nimh.nih.gov

Dear experts,

I am doing a resting state project using older adults, MCI, and AD patients. For the project, I am performing a series of analyses, of which the input to each analysis is a symmetric matrix of functional connectivity. To form this matrix, I will extract the mean time series from a number of ROIs and then Pearson correlate.

Given the population, I am having some trouble with motion censoring during preprocessing. For my preprocessing pipeline, I am following example 11 from the proc.py help page:

afni_proc.py -subj_id $sid -script proc.$sid -scr_overwrite \
-blocks despike tshift align tlrc volreg mask blur scale \
regress \
-radial_correlate_blocks tcat volreg \
-copy_anat $anat_ss \
-anat_has_skull no \
-anat_follower anat_w_skull anat $anat_orig \
-anat_follower_ROI aaseg anat $aparc_anat \
-anat_follower_ROI aeseg epi $aparc_anat \
-anat_follower_ROI FSvent epi $vent_roi \
-anat_follower_ROI FSWe epi $wm_roi \
-anat_follower_erode FSvent FSWe \
-dsets $EPI_files \
-tcat_remove_first_trs $dummy_scans \
-tcat_remove_last_trs $rm_end_TRs \
-align_opts_aea -cost lpc+ZZ -$move \
-check_flip \
-tlrc_base MNI152_2009_template_SSW.nii.gz \
-tlrc_NL_warp \
-tlrc_NL_warped_dsets $anat_std $anat_warp_aff \
$anat_warp_nl \
-volreg_align_to MIN_OUTLIER \
-volreg_align_e2a \
-volreg_tlrc_warp \
-blur_to_fwhm \
-blur_size $blur_size \
-mask_epi_anat yes \
-regress_motion_per_run \
-regress_ROI_PC FSvent 3 \
-regress_ROI_PC_per_run FSvent \
-regress_make_corr_vols aeseg FSvent \
-regress_anaticor_fast \
-regress_anaticor_label FSWe \
-regress_censor_motion 0.3 \
-regress_censor_outliers 0.10 \
-regress_apply_mot_types demean deriv \
-regress_est_blur_epits \
-regress_est_blur_errts \
-html_review_style pythonic


Based on the literature, I tweaked the motion censoring threshold to 0.3 mm and the outlier threshold to 0.10 to better fit my population. While the pipeline is able to run through successfully for each participant, in a majority of my sample, over 10% of TRs are being censored. The problem with this is that when I extract the average time series from the errts dataset for each ROI, TRs that were censored have a value of 0 in the time series matrix (see attached picture). This is falsely inflating the correlation between ROIs since 0s are present systematically throughout each participant’s time series x ROI matrix. Do you have any suggestions to get around this? Is it reasonable to use a higher motion threshold so that fewer TRs are censored? If I make the threshold too lenient will my correlation matrix be similarly affected due to motion?

Thank you for your help!
Jenna Blujus