Thanks, the data makes it very clear.

The problem may stem from the fact that there were
only 21 time points used for the correlations, so the
r -> p-value conversion does not lead to the small
p-values you may be used to with 150 TR time series
data, for example.

Looking at the afni controller, it is not just the q-values
that are big, it is also the p-values. And the distribution
of these correlation values ends up being no better than
random chance, which is why the q-values do not go
much lower than 1.

Basically, set the p-value to 0.01 or 0.001, for example.
In order to get a good q-value, there must be a high
fraction of the masked volume of voxels that survive
the threshold. There aren't enough voxels surviving.

- rick