Correct way to extract mean seed-driven functional connectivity networks from 3dLMEr model

AFNI Message Board
1

Hello Gang et al.
I am running a rather simple within subjects LMM on resting state fMRI scans of 50 subjects who experienced 2 different scans, which I label ‘states’.

When I attempt to use 3dmerge to extract the mean state-dependent functional connectivity maps from the gltCode (using, e.g : 3dmerge -1clust 3 54 -1thresh 7.14 -1erode 50 -1dilate -prefix newfilenamep10-12 ‘outputfilefrom3dlmer+tlrc[4]’)

The Z-stats in the 4th subbrik are astronomical (z-stat 7.14 corresponds to a threshold of p = 1x10^-12) and the maps still cover quite a lot of the brain at this level. The thresholded maps are attached for reference (seed-driven FC from the left amygdala for each state).

  1. Are these statistics valid? Even between states within the same participants within the same scanning session, the appearance of a difference is very dramatic.
  2. How would I arrive at a rational statistical threshold for the mean functional connecitivity maps? the p-value of these maps is 1x10^-12 and the cluster extent is 54 mm^3 (2 voxels in real space).
  3. or did I make a serious mistake somewhere?

Thank you.

Tim Meeker

The full model specification is below:

for seed in bilataMCC #bilatpgACC bilatsgACC bilatspgACC PAGanat
do

3dLMEr -prefix Capsaicin${seed}gm -jobs 8
-mask /data/tim/DPMN2019/gncREST_groupmask+tlrc
-model ‘state+(1|Subj)’
-gltCode control ‘state : 1control’
-gltCode pain 'state : 1pain’
-gltCode pain-control ‘state : 1pain -1control’
-dataTable
Subj state InputFile
s1 control /data/tim/DPMN2019/1stlevel/{seed}gnc1007REST1Z_gm+tlrc \ s1 pain /data/tim/DPMN2019/1stlevel/{seed}gnc1007REST2Z_gm+tlrc \
s2 control /data/tim/DPMN2019/1stlevel/{seed}gnc1022REST1Z_gm+tlrc \ s2 pain /data/tim/DPMN2019/1stlevel/{seed}gnc1022REST2Z_gm+tlrc \
s3 control /data/tim/DPMN2019/1stlevel/{seed}gnc1048REST1Z_gm+tlrc \ s3 pain /data/tim/DPMN2019/1stlevel/{seed}gnc1048REST2Z_gm+tlrc \
s4 control /data/tim/DPMN2019/1stlevel/{seed}gnc1054REST1Z_gm+tlrc \ s4 pain /data/tim/DPMN2019/1stlevel/{seed}gnc1054REST2Z_gm+tlrc \
s5 control /data/tim/DPMN2019/1stlevel/{seed}gnc1060REST1Z_gm+tlrc \ s5 pain /data/tim/DPMN2019/1stlevel/{seed}gnc1060REST2Z_gm+tlrc \
s6 control /data/tim/DPMN2019/1stlevel/{seed}gnc1062REST1Z_gm+tlrc \ s6 pain /data/tim/DPMN2019/1stlevel/{seed}gnc1062REST2Z_gm+tlrc \
s7 control /data/tim/DPMN2019/1stlevel/{seed}gnc1088REST1Z_gm+tlrc \ s7 pain /data/tim/DPMN2019/1stlevel/{seed}gnc1088REST2Z_gm+tlrc \
s8 control /data/tim/DPMN2019/1stlevel/{seed}gnc2003REST1Z_gm+tlrc \ s8 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2003REST2Z_gm+tlrc \
s9 control /data/tim/DPMN2019/1stlevel/{seed}gnc2005REST1Z_gm+tlrc \ s9 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2005REST2Z_gm+tlrc \
s10 control /data/tim/DPMN2019/1stlevel/{seed}gnc2017REST1Z_gm+tlrc \ s10 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2017REST2Z_gm+tlrc \
s11 control /data/tim/DPMN2019/1stlevel/{seed}gnc2029REST1Z_gm+tlrc \ s11 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2029REST2Z_gm+tlrc \
s12 control /data/tim/DPMN2019/1stlevel/{seed}gnc2032REST1Z_gm+tlrc \ s12 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2032REST2Z_gm+tlrc \
s13 control /data/tim/DPMN2019/1stlevel/{seed}gnc2044REST1Z_gm+tlrc \ s13 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2044REST2Z_gm+tlrc \
s14 control /data/tim/DPMN2019/1stlevel/{seed}gnc2053REST1Z_gm+tlrc \ s14 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2053REST2Z_gm+tlrc \
s15 control /data/tim/DPMN2019/1stlevel/{seed}gnc2056REST1Z_gm+tlrc \ s15 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2056REST2Z_gm+tlrc \
s16 control /data/tim/DPMN2019/1stlevel/{seed}gnc2057REST1Z_gm+tlrc \ s16 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2057REST2Z_gm+tlrc \
s17 control /data/tim/DPMN2019/1stlevel/{seed}gnc2058REST1Z_gm+tlrc \ s17 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2058REST2Z_gm+tlrc \
s18 control /data/tim/DPMN2019/1stlevel/{seed}gnc2063REST1Z_gm+tlrc \ s18 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2063REST2Z_gm+tlrc \
s19 control /data/tim/DPMN2019/1stlevel/{seed}gnc2065REST1Z_gm+tlrc \ s19 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2065REST2Z_gm+tlrc \
s20 control /data/tim/DPMN2019/1stlevel/{seed}gnc2066REST1Z_gm+tlrc \ s20 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2066REST2Z_gm+tlrc \
s21 control /data/tim/DPMN2019/1stlevel/{seed}gnc2071REST1Z_gm+tlrc \ s21 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2071REST2Z_gm+tlrc \
s22 control /data/tim/DPMN2019/1stlevel/{seed}gnc2072REST1Z_gm+tlrc \ s22 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2072REST2Z_gm+tlrc \
s23 control /data/tim/DPMN2019/1stlevel/{seed}gnc2074REST1Z_gm+tlrc \ s23 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2074REST2Z_gm+tlrc \
s24 control /data/tim/DPMN2019/1stlevel/{seed}gnc2075REST1Z_gm+tlrc \ s24 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2075REST2Z_gm+tlrc \
s25 control /data/tim/DPMN2019/1stlevel/{seed}gnc2077REST1Z_gm+tlrc \ s25 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2077REST2Z_gm+tlrc \
s26 control /data/tim/DPMN2019/1stlevel/{seed}gnc2079REST1Z_gm+tlrc \ s26 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2079REST2Z_gm+tlrc \
s27 control /data/tim/DPMN2019/1stlevel/{seed}gnc2083REST1Z_gm+tlrc \ s27 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2083REST2Z_gm+tlrc \
s28 control /data/tim/DPMN2019/1stlevel/{seed}gnc2084REST1Z_gm+tlrc \ s28 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2084REST2Z_gm+tlrc \
s29 control /data/tim/DPMN2019/1stlevel/{seed}gnc2085REST1Z_gm+tlrc \ s29 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2085REST2Z_gm+tlrc \
s30 control /data/tim/DPMN2019/1stlevel/{seed}gnc2087REST1Z_gm+tlrc \ s30 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2087REST2Z_gm+tlrc \
s31 control /data/tim/DPMN2019/1stlevel/{seed}gnc2089REST1Z_gm+tlrc \ s31 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2089REST2Z_gm+tlrc \
s32 control /data/tim/DPMN2019/1stlevel/{seed}gnc2093REST1Z_gm+tlrc \ s32 pain /data/tim/DPMN2019/1stlevel/{seed}gnc2093REST2Z_gm+tlrc \
s33 control /data/tim/DPMN2019/1stlevel/{seed}gnc3004REST1Z_gm+tlrc \ s33 pain /data/tim/DPMN2019/1stlevel/{seed}gnc3004REST2Z_gm+tlrc \
s34 control /data/tim/DPMN2019/1stlevel/{seed}gnc3006REST1Z_gm+tlrc \ s34 pain /data/tim/DPMN2019/1stlevel/{seed}gnc3006REST2Z_gm+tlrc \
s35 control /data/tim/DPMN2019/1stlevel/{seed}gnc3009REST1Z_gm+tlrc \ s35 pain /data/tim/DPMN2019/1stlevel/{seed}gnc3009REST2Z_gm+tlrc \
s36 control /data/tim/DPMN2019/1stlevel/{seed}gnc3010REST1Z_gm+tlrc \ s36 pain /data/tim/DPMN2019/1stlevel/{seed}gnc3010REST2Z_gm+tlrc \
s37 control /data/tim/DPMN2019/1stlevel/{seed}gnc3011REST1Z_gm+tlrc \ s37 pain /data/tim/DPMN2019/1stlevel/{seed}gnc3011REST2Z_gm+tlrc \
s38 control /data/tim/DPMN2019/1stlevel/{seed}gnc3014REST1Z_gm+tlrc \ s38 pain /data/tim/DPMN2019/1stlevel/{seed}gnc3014REST2Z_gm+tlrc \
s39 control /data/tim/DPMN2019/1stlevel/{seed}gnc3016REST1Z_gm+tlrc \ s39 pain /data/tim/DPMN2019/1stlevel/{seed}gnc3016REST2Z_gm+tlrc \
s40 control /data/tim/DPMN2019/1stlevel/{seed}gnc3017REST1Z_gm+tlrc \ s40 pain /data/tim/DPMN2019/1stlevel/{seed}gnc3017REST2Z_gm+tlrc \
s41 control /data/tim/DPMN2019/1stlevel/{seed}gnc3018REST1Z_gm+tlrc \ s41 pain /data/tim/DPMN2019/1stlevel/{seed}gnc3018REST2Z_gm+tlrc \
s42 control /data/tim/DPMN2019/1stlevel/{seed}gnc3019REST1Z_gm+tlrc \ s42 pain /data/tim/DPMN2019/1stlevel/{seed}gnc3019REST2Z_gm+tlrc \
s43 control /data/tim/DPMN2019/1stlevel/{seed}gnc3021REST1Z_gm+tlrc \ s43 pain /data/tim/DPMN2019/1stlevel/{seed}gnc3021REST2Z_gm+tlrc \
s44 control /data/tim/DPMN2019/1stlevel/{seed}gnc3023REST1Z_gm+tlrc \ s44 pain /data/tim/DPMN2019/1stlevel/{seed}gnc3023REST2Z_gm+tlrc \
s45 control /data/tim/DPMN2019/1stlevel/{seed}gnc3024REST1Z_gm+tlrc \ s45 pain /data/tim/DPMN2019/1stlevel/{seed}gnc3024REST2Z_gm+tlrc \
s46 control /data/tim/DPMN2019/1stlevel/{seed}gnc3025REST1Z_gm+tlrc \ s46 pain /data/tim/DPMN2019/1stlevel/{seed}gnc3025REST2Z_gm+tlrc \
s47 control /data/tim/DPMN2019/1stlevel/{seed}gnc3030REST1Z_gm+tlrc \ s47 pain /data/tim/DPMN2019/1stlevel/{seed}gnc3030REST2Z_gm+tlrc \
s48 control /data/tim/DPMN2019/1stlevel/{seed}gnc3031REST1Z_gm+tlrc \ s48 pain /data/tim/DPMN2019/1stlevel/{seed}gnc3031REST2Z_gm+tlrc \
s49 control /data/tim/DPMN2019/1stlevel/{seed}gnc3034REST1Z_gm+tlrc \ s49 pain /data/tim/DPMN2019/1stlevel/{seed}gnc3034REST2Z_gm+tlrc \
s50 control /data/tim/DPMN2019/1stlevel/{seed}gnc3035REST1Z_gm+tlrc \ s50 pain /data/tim/DPMN2019/1stlevel/{seed}gnc3035REST2Z_gm+tlrc \
done

leftamygdalapainmeanp10-12z=-30-26-22-18-14-10-6-2.jpg
leftamygdalapainmeanp10-12z=-30-26-22-18-14-10-6-2.jpg1200×716 75.6 KB

leftamygdalacontrolmeanp10-12z=-30-26-22-18-14-10-6-2.jpg
leftamygdalacontrolmeanp10-12z=-30-26-22-18-14-10-6-2.jpg1206×716 73.5 KB

2

Hi Tim,

Are these statistics valid?

You seem to have a simple case of traditional paired t-test. You could verify the 3dLMEr result with, for example, 3dttest++.

How would I arrive at a rational statistical threshold for the mean functional connecitivity maps? the p-value of these
maps is 1x10^-12 and the cluster extent is 54 mm^3 (2 voxels in real space).

Rationality is in the eye of beholder! I don’t take the voxel-wise p-value and the cluster threshold too seriously. If possible, a region-based approach might be more preferable at least for two advantages: 1) avoid spatial boundary ambiguities, and 2) focus on effect estimation instead of dichotomania.

3

Are these statistics valid?

You seem to have a simple case of traditional paired t-test. You could verify the 3dLMEr result with, for example, 3dttest++.

OK. I have those results anyway I think…

How would I arrive at a rational statistical threshold for the mean functional connecitivity maps? the p-value of these
maps is 1x10^-12 and the cluster extent is 54 mm^3 (2 voxels in real space).

Rationality is in the eye of beholder! I don’t take the voxel-wise p-value and the cluster threshold too seriously. If possible, a region-based approach might be more preferable at least for two advantages: 1) avoid spatial boundary ambiguities, and 2) focus on effect estimation instead of dichotomania.

From the standpoint of reality, I absolutely agree with you. From the standpoint of convincing users of Some other PrograM, or Connectivity for our Other Nearest Neighbors I am running into friction.

Perhaps I will try an R-map approach. in addition to a dichotomania approach.

Thanks,
Tim

4

Also, curious where would I find the most accurate estimation of the degrees of freedom remaining in each dataset (after preprocessing)? And across individuals how would this scale? I’m assuming it would not scale linearly or anything like that.

5

I’m not aware of an approach to taking into consideration the degrees of freedom after preprocessing when computing the seed-based correlation at the subject level. It remains opaque as to how much this would impact at subject and population level.

6

Neither am I, which is why it remains an interesting question I guess. I’ll tinker with some things.