Any of these steps can be detrimental to the others (though
maybe despiking doesn't hurt time shifting so much). And
there is no method that is considered "best", we do what
we believe is most reasonable. What might work well for
one person's data might not work well for another (such as
if some subjects are very prone to movement).

So you don't need to worry about a word like "contradictory",
since we know of no mathematical way to say one way is best.

----

I can imagine that despiking might cause a small difficulty
in volume registration. However, a "spike" suggests quick
movements. And not just one movement, but multiple.

Multiple quick movements make volume registration useless,
since different slices of the dataset will probably have
vastly different registration requirements. Which is to say
that 3D volume registration cannot succeed (at that time).

{Some people try 2D registration (sagittal slices only) for
that reason, but I think you have to be pretty luck for it
to work well.}

That is one reason that people censor the TRs with spikes.
Such motion is uncorrectable.

----

No, time shifting should not help registration, the goals
are different. It's more a question of whether (and how
much) it hurts.

The only "beneficial" step that I can see is that despiking
might be helpful for time shifting (since BOLD signals are
smooth and slow, there should not be any spikes).

----

If motion is a very significant problem for you, consider
adding a gentle real-time feed back mechanism for your
subjects, so they can literally see when they are moving
"too much". Some people use small red/green lights. The
path of:

Dimon -> afni realtime plugin -> serial_helper

has the ability to help send data back to the subjects.
The scanners here at NIH are setup to do that easily.

Simply asking your subjects to move less is only a little
helpful. Having a visual cue can make a big difference.

- rick