mask for statistical analysis

Dear all,

In the statistical analysis, we need to choose a binary mask.

What kind of binary masks I can apply? If I turn to the group-wise GM mask, it seems there're too many comparisons, leading to failure in the tfce+fwe correction (p<0.05).

Can I use the |Z|>2 of the Z map as the binary mask instead to reduce the number of comparisons?

Thanks!