Blur resting state before ETAC?


Two quick questions:

  1. I’m following referencing example 11 to process resting state data. I plan on doing group-level analyses with ETAC - I know that ETAC should do my blurring for me at various levels, so should I still blur the RS data as part of the pre-processing?

  2. For task-based pipelines, I’d like to extract mean beta coefficients following ETAC output in order to report parameter estimates. Given that some clusters are only going to be detected at certain blur levels my current pipeline is to mine the ETACmaskALL output to (a) collapse across all p-values for each blur, (b) extract those clusters, (c) blur the deconvolved data with the various blurs, (d) and then match up the clusters and blurred deconvolved data appropriately to extract mean beta coefficients. Then I’ll average the mean betas across all blurs for reporting. Is this a reasonable approach?

# afni script used for question number 1 -subj_id sub-3156                            \
  -blocks despike tshift align tlrc volreg blur mask         \
		  scale regress                                      \
  -copy_anat sub-3156/struct+orig                            \
  -dsets sub-3156/Resting+orig.HEAD                          \
  -tcat_remove_first_trs 0                                   \
  -align_opts_aea -cost lpc+ZZ                               \
  -tlrc_base vold2_mni_brain+tlrc                            \
  -tlrc_NL_warp                                              \
  -volreg_align_to MIN_OUTLIER                               \
  -volreg_align_e2a                                          \
  -volreg_tlrc_warp                                          \
  -volreg_warp_dxyz 1.8                                      \
  -mask_segment_anat yes                                     \
  -mask_segment_erode yes                                    \
  -mask_import Tvent final_mask_CSF+tlrc 			         \
  -mask_intersect Svent CSFe Tvent                           \
  -mask_epi_anat yes                                         \
  -regress_motion_per_run                                    \
  -regress_ROI_PC Svent 3                                    \
  -regress_ROI_PC_per_run Svent                              \
  -regress_make_corr_vols WMe Svent                          \
  -regress_anaticor_fast                                     \
  -regress_censor_motion 0.2                                 \
  -regress_censor_outliers 0.05                              \
  -regress_apply_mot_types demean deriv                      \
  -regress_est_blur_epits                                    \
  -regress_est_blur_errts                                    \
  -regress_run_clustsim yes


### A quick sample of some code used in question number 2

blurArr=(4 6 8); numBlur=${#blurArr[@]}
pval_list=(0.01 0.005 0.001); numPval=${#pval_list[@]}

### Mine ETACmaskALL
blurC=0; pvalC=0
while [ $blurC -lt $numBlur ]; do

	# build list of p-values for each blur
	unset hBrick
	for ((j=1; j<=$numPval; j++)); do
		let pvalC=$[$pvalC+1]

	# extract all sub-bricks for each blur
	3dbucket -prefix FINAL_b${blurArr[$blurC]}_${out} \

	# collapse across all p-values in each blur
	3dmask_tool -input FINAL_b${blurArr[$blurC]}_${out}+tlrc \
	-union -prefix FINAL_b${blurArr[$blurC]}_allP_${out}

	let blurC=$[$blurC+1]

### Extract clusters
for i in FINAL_*allP*.HEAD; do

	tmp1=${i%_ET*}; pref=${tmp1##*_}
	tmp2=${i#*_}; blur=${tmp2%%_*}

	3dclust -1Dformat -nosum -1dindex 0 \
	-1tindex 0 -2thresh -0.5 0.5 -dxyz=1 \
	-savemask Clust_${pref}_${blur}_mask \
	1.01 5 $i > Clust_${pref}_${blur}_table.txt

### pull betas from appropriately blurred data for each mask
subjList=(array of subjects)

# loop through blurs
d=0; while [ $d -lt $numBlur ]; do

	# determine number of clusters in each blur mask
	3dcopy Clust_SpT1_b${blurInt}_mask+tlrc SpT1_b${blurInt}.nii.gz

	num=`3dinfo Clust_SpT1_b${blurInt}_mask+tlrc | \
	grep "At sub-brick #0 '#0' datum type is short" | \
	sed 's/[^0-9]*//g' | sed 's/^...//'`

	# split clust masks into individual files
	for (( j=1; j<=$num; j++ )); do
		c3d SpT1_b${blurInt}.nii.gz -thresh $j $j 1 0 -o SpT1_b${blurInt}_${j}.nii.gz
		3dcopy SpT1_b${blurInt}_${j}.nii.gz Clust_SpT1_b${blurInt}_c${j}+tlrc

	# pull betas
	for i in Clust_SpT1_b${blurInt}_c*+tlrc.HEAD; do

		tmp=${i##*_}; cnum=${tmp%+*}
		> $print

		for j in ${subjList[@]}; do


			# blur decon data
			3dmerge -prefix ${subjDir}/SpT1_decon_REML_blur${blurInt} \
			-1blur_fwhm $blurInt -doall ${subjDir}/SpT1_decon_REML+tlrc

			stats=`3dROIstats -mask $i "${file}[${betas}]"`
			echo "$j $stats" >> $print
	let d=$[$d+1]


Sorry for being slow. This was before our week-long AFNI bootcamp…

  1. If you choose to use ETAC for blurring, then it is preferable not to blur at the individual level. We did an analysis like that, and then had ETAC blur at levels 0,2,4,6,8 (but perhaps 8 is getting too big, it depends on the voxel size). That is a very reasonable way to go.

  2. ETAC will output betas/t-stats at each blur level, in addition to the final voxels of significance. You might get cluster averages out of the non-blurred results. The average should be pretty robust, and the non-blurred one will have the least contamination from outside the cluster. A cluster average is like a large blur within the ROI.

Yeah, your script for this is messy. But I don’t see a good reason to use blurred data. The results should not change much.

  • rick


Thank you for the response, I really appreciate your input.