Creating conjunction map w/ two different clustering methods

AFNI Message Board
1

Hello,

Within my analysis, I have two different models that I would like to compare using a conjunction map. The first is an F map from 3dLME and the second is the result from 3dttest++. Currently, I am using clustsim and 3dClusterize to compute the F map in 3dLME at p < 0.001 a=0.05. For 3dttest++ I have used the following ETAC flag:
-ETAC_optNN=3:side=1,2:hpow=0,1,2:pthr=0.02,0.005,0.001:fpr=ALL:name=TEST \

My broader question is, is it appropriate to compare these two maps despite the different clustering methods?

I also have several questions about the ETAC output. It seems to be more liberal compared to clustsim/3dClusterize method. Is this due to the lower p threshold included in the ETAC command? In addition, the resulting map has two large clusters labeled as 4 and 6. I am not quite sure how to interpret these cluster numbers as they are not sequential. Do 4 and 6 have a specific meaning? I also ran 3dclust on this output to see if I could parse apart the two big clusters and there were 35 resulting clusters which seems large. Is this an appropriate practice?

2

I have two different models that I would like to compare using a conjunction map

A conjunction map would misrepresent/distort the situation due to the artificial dichotomization process. I would rather trust your eyeball assessment than such a conjunction map.

is it appropriate to compare these two maps despite the different clustering methods? … It seems to be more liberal compared to clustsim/3dClusterize method. Is this due to the lower p threshold included in the ETAC command?

Each adjustment approach has some underlying assumptions, but most of the time (but not always) the ETAC approach is more adaptive and flexible with a much wider range of statistical evidence as opposed to the cluster-based methods in which the statistical evidence is bounded by an upper limit.

I also ran 3dclust on this output to see if I could parse apart the two big clusters and there were 35 resulting clusters which seems large. Is this an appropriate practice?

This concerns one big problem of the conventional multiple testing adjustment approaches (in addition to many other issues): the only consideration is about statistical evidence lumping without regard to any anatomical information.