How to get the Roi of ConditionA-mean(ConditionB+ConditionC+ConditionD+ConditionE) ?

AFNI Message Board
1

I come up with two method,
one is using the 3dANOVA with option: -levels 5 , -DAFNI_FLOATIZE=YES and -contr 4 -1 -1 -1 -1, with every subject’s coef value from results of 3dDeconvolve
the other is using 3dttest++ with opiton: -setA, followed by every subject’s data corresponding to option -gltsym A -0.25B -0.25C -0.25D -0.25E of 3dDeconvolve.
the difference betwwen those two method.
I want to know which one is right and how to set the P value to get the cluster(ROI)

the following is part of my code

3dDeconvolve -input pb06.$subj.r01.scale+tlrc.HEAD
pb06.$subj.r02.scale+tlrc.HEAD pb06.$subj.r03.scale+tlrc.HEAD
…
-num_stimts 17
-stim_times 1 stimuli/AV1_A_halfo.txt ‘BLOCK(12,1)’
-stim_label 1 snake
-stim_times 2 stimuli/AV2_B_halfo.txt ‘BLOCK(12,1)’
-stim_label 2 mushroom
-stim_times 3 stimuli/AV3_C_halfo.txt ‘BLOCK(12,1)’
-stim_label 3 spider
-stim_times 4 stimuli/AV4_D_halfo.txt ‘BLOCK(12,1)’
-stim_label 4 animal
-stim_times 5 stimuli/AV5_E_halfo.txt ‘BLOCK(12,1)’
-stim_label 5 face
…
-gltsym ‘SYM: 0.2A +0.2B +0.2C +0.2D +0.2E’
-glt_label 12 meanABCDE
-gltsym 'SYM: A -0.25B -0.25C -0.25D -0.25*E’
-glt_label 13 A-mean(BCDE)
…
-errts errts.${subj}.$cond
-bucket stats.$subj.$cond

tcsh -x stats.REML_cmd

3dANOVA -levels 5 -DAFNI_FLOATIZE=YES
-dset 1 ‘stats.S01.new_REML+tlrc[1]’
-dset 1 ‘stats.S02.new_REML+tlrc[1]’
-dset 1 ‘stats.S03.new_REML+tlrc[1]’
…
dset 5 ‘stats.S08.new_REML+tlrc[13]’
-dset 5 ‘stats.S09.new_REML+tlrc[13]’
-dset 5 ‘stats.S10.new_REML+tlrc[13]’
-ftr stim
-mean 1 A
-mean 2 B
-mean 3 C
-mean 4 D
-mean 5 E
-contr 4 －1 －1 －1 －1 A-mean(BCDE)
-contr 1 1 1 1 1 ABCD \

3dttest++ -prefix A-mean(BCDE)
-setA ‘stats.S01.new_REML+tlrc[52]’
‘stats.S02.new_REML+tlrc[52]’
‘stats.S03.new_REML+tlrc[52]’
‘stats.S04.new_REML+tlrc[52]’
‘stats.S05.new_REML+tlrc[52]’
‘stats.S06.new_REML+tlrc[52]’
‘stats.S07.new_REML+tlrc[52]’
‘stats.S08.new_REML+tlrc[52]’
‘stats.S09.new_REML+tlrc[52]’
‘stats.S10.new_REML+tlrc[52]’
-overwrite

2

Hi Shawn,

I will ignore how to set the p-value, since it is not
clear what you want to set it based on (plus, it should
be easy to set in the GUI).

What are you seeing as differences between those 2
tests? From what I can tell, they should be identical,
except that the ANOVA version should be exactly 4 times
as big. Note that the 3dDeconvolve/3dttest++ version
seems more reasonable to me, where each side of the
contrast sums to 1.0.

But even with the 4 times scaling in the ANOVA version,
the t-stats should be identical. Is that what you are
seeing?

  • rick
3

The 3dttest++ approach is fine. However, the 3dANOVA method is not correct; instead, you should use 3dANOVA2 -type 3 with weights of -acontr 1 -0.25 -0.25 -0.25 -0.25. Keep in mind that 3dANOVA is for one-way between-subjects ANOVA while 3dANOVA2 -type 3 is for one-way within-subject ANOVA.

4

Oh, I just glossed over the use of 3dANOVA vs 3dANOVA2, and was
picturing the latter. So the 3dANOVA command would be as if there
were 5 groups, say, using that contrast, is that right?

Thanks,

  • rick
5

So the 3dANOVA command would be as if there
were 5 groups, say, using that contrast, is that right?

Yes, exactly.

6

thanks very much

7

Dear expert
My new question is related to the old question. I used 3dttest++ with option: -setA conditionA_beta, -setB mean(condition B,C,D,E)_beta -paird, and -Clustsim. Then I get the threshold files named ‘cc.XXX’.

    I also used 3dFWHMx with option -ACF -input errts_REML+tlrc, which is the output of '3dDeconvolve -errts' from one subject's data. Then I run 3dClustsim with ACF values from 3dFWHMx and get a different threshold table.

    But What I confused is that the  threshold from 3dttest++ is much stricter than threshold from results combining 3dFWHMx and 3dClustsim.

    I can get Roi saved at erea V1 if I used the latter method. However I get nothing at the whole brain if I use the first method.

    which threshold I should use. Can you give some remendation.

thanks very much.

8

I can get Roi saved at erea V1 if I used the latter method. However I get nothing at the whole brain if I use the first method.

It’s hard to know what’s going on without the data. The underlying mechanism for the two approaches is slightly different. See the following teaching material for details: https://afni.nimh.nih.gov/pub/dist/edu/latest/afni_handouts/Clusters_2017.pdf