Dear John,
> as long as coregistration between the functional and structural scans is good, fieldmap correction of EPIs may not be necessary
Well, the geometric shape of the functional data is always going to differ from that of the structure to some extent. As the default coregistration is just based on rigid-body transformations the quality of the coregistration will always be limited (in an extreme case, think of trying to register an ellipsoid onto a sphere). The displacement will also vary between regions, e.g. the distortions might be especially large in the most anterior and posterior regions of the brain, which might be very problematic if you're interested e.g. in retinotopy and surface-based analyses (see Vasseur et al., 2010, J Vis, https://dx.doi.org/10.1167/10.12.30 ). This would hold for any normalisation strategy. A better indirect normalisation (via T1) will only increase registration accuracy between subjects for the T1s, but not affect alignment between T1 and EPI within subjects.
Sometimes people just go with direct normalisation onto the EPI template when encountering large artefacts in the EPI data. This only solves the issue at first sight though, as the implicit assumption is that their data has deformations/drop-outs very similar to the EPI template, which probably is not the case. If signal in OFC is completely lost then more dorsal frontal regions might be warped downward, with normalised data looking "good", but this doesn't mean the signal can be interpreted as OFC.
Thus usually, a field map is a good idea, especially as its acquistion doesn't take long, but as stated, it's tricky if subjects move between field map and EPI session (or if there's relatively large motion within the session). Other strategies might be more promising, e.g. two EPI sessions with reversed blip direction, see e.g. http://www.diffusiontools.com/documentation/hysco.html . Alternatively, just hope that distortions are relatively small and similar across subjects.
Best
Helmut
|