General question about alignment (epi2anat vs anat2epi)

I have always been confused if it’s more appropriate to align the epi to the anat and then run the rest of the preprocessing steps on the newly aligned epi file, or to align the ant to the epi and run the preprocessing steps on the original epi file.

Does anyone have advice or can provide a brief reasoning why I shouldn’t do one or the other?

This becomes more confusing because I would like to bring the participant space to MNI space using 3dNwarpApply. I realize that if I chose epi2ant then I can just use the original anat space. Otherwise, if I chose anat2epi, then I would have to use the newly aligned anat file, correct?


The differences in the actual results will likely be subtle. With anat2epi, there is one more transformation for the anatomical dataset. Also, the alignment of the anatomical dataset to a template (via @auto_tlrc) is based on the interpolated anat dataset. The final anatomical dataset will include all transformations including that anat to epi transformation. With the epi2anat alignment, the epi dataset gets the extra alignment transformation, and the anatomical dataset’s set of transformations is simpler by one transformation. These transformations are combined (cat_matvec) an into 3dNwarpApply/3dAllineate. You can tell the differences by running with and without the -volreg_align_e2a option. Compare the proc script output for the alignments and the way they use the transformations.