Is it needed to use -mask_apply epi option for subject-individual-space analysis?

AFNI Message Board
1

Hi,

afni_proc helping file says mask_apply epi option is not default since 2009, and recommends using same group mask in latter group analysis, if I understand correctly. But is it correct or needed to use -mask_apply epi option for subject-individual-space, not group, no tlrc block, analysis?

Thank you,
2086

2

Hi 2086,

So you will be reporting individual subject results in original space?

The time to apply a mask is when correcting for multiple comparisons/
tests. If that is in orig space, then yes, an orig space mask is needed.
But still, such a mask need not be applied until actually clustering or
otherwise correcting the results. The orig space linear regression need
not be masked, for example.

  • rick
3

Hi Rick,

Thanks for your reply.

So you will be reporting individual subject results in original space?

Yes, we are going to get ROIs in original space for other analysis: 1, extract information in these ROIs, then do some correlation and group comparison; 2, DTI tractography, which needs precise ROI location.

But still, such a mask need not be applied until actually clustering or otherwise correcting the results. The orig space linear regression need not be masked, for example.

I didn’t fully understand what you mean. The differences between proc scripts produced by afni_proc with and without mask_apply epi option are:
1, In the scale step, full_mask vs mask_extents;
2, In the regress step, full_mask vs no mask.
In the later blur estimation step, both proc scripts use full_mask for correcting for multiple comparisons.

Thank you again.
2086

4

Hi 2086,

If you plan to do an ROI analysis, then assuming the ROIs are actually inside the brain masks, then applying the brain mask won’t make any difference.

Both the scaling and regress steps use a mask to decide which voxels to zero out. Other voxels are unaffected. That is why you need not apply a mask, it only hides results. One would usually prefer to see more results.

  • rick
5

Hi Rick,

Thank you very much!

2086