Dear Paul,
As this might be interesting for future versions: In case of fast head motion the overall shape of the brain would be "distorted" as well due to spatially "shifted" slices within the volume, accomplishing the rigid-body registration in SPM during realignment, and affecting the rp estimates and the mean image. It might be worth a try to go with realigned data, determine the bad volumes, discard them, rerun realign on the raw data minus the bad volumes (if one wants to realign on the mean image this should be a little more precise, as the mean image is based on "good" data) and add interpolated volumes plus corresponding values into the rp file where necessary. This way one would avoid abrupt jerks stored in the rp file although the corresponding volumes have been removed from the data by adding interpolated volumes (thus basically trying to predict something that has already been removed from the data series).
I'm aware of additional options in Artrepair, e.g. the slice-wise correction, which might already compensate well for displacement of slices within volumes. But the approach as described above *could* also be useful.
Best
Helmut
|