Dear Daniel and Patricia,
a few thoughts in the meantime: Basically it boils down to the issue whether rigid-body transformations are sufficient to explain the displacement between runs/sessions or not. If subjects undergo several runs within a single fMRI session motion is (hopefully) rather small. In that case geometric distortions, signal loss, ... between the different runs should be quite similar (at least when going with the same scanning parameters for different EPI runs). Then it should be more precise to realign / coregister within modality, so EPI session onto EPI session, structure onto structure. Due to its nature the between-modality coregistration (structure onto EPI), the rigid-body transformation part of the normalisation or segmentation (structure onto SPM templates) and the warping of the normalisation can be expected to be more prone to errors. If you realign/coregister everything within modality first, then all the EPI sessions undergo the same good or bad coregistration to the average structure and the same good or bad normalisation though, so the whole data set should be biased the same way.
However, if the displacement is several cm then it is quite likely that geometric distortions, signal loss, ... of the different runs vary to some noticeable extent. It should be more precise then to non-rigidly warp the different individual data sets onto the individual's average. In contrast, separately warping the different runs of a particular subject into MNI space can be assumed to be suboptimal, as the difference between different runs is probably still much smaller compared to the difference of a specific run and any of the templates.
In the end, this leads to the question at which point the rigid-body transformations are not sufficient any longer. And unfortunately there's no final answer, as it's going to depend on the data (could also be a nice methods paper ;) As functional data undergo smoothing (thus somewhat correcting for remaining normalisation errors to some extent) existing differences between EPI runs might be noticable in the raw data but at the same time "too small" / irrelevant for analysis.
Best,
Helmut
|