Dear Balu,
when conducting single-subject analyses the data has to be resliced at some point (this is done during normalisation step otherwise). As Marko stated, during "Realignment: Estimate" the headers of the images are adjusted, but the data itself is not resliced (motion parameters are not applied to the data). This means one voxel in volume 1 does not correspond to the same voxel in volume 2. However, during slice timing motion parameters are NOT applied to the data either (nor during smoothing). To solve this issue you have to reslice your data somewhere. Otherwise you get some error similar to yours when trying to run single-subject models. The same error might occur during group statistics if data from different subjects weren't normalised the same way (something like different bounding box) or if you try to run group statistics on data in native space.
Concerning the order of slice timing and realignment see http://mindhive.mit.edu/node/109 for more information. With interleaved acquisiton scheme it is probably better to start with slice timing [if you expect rather large motion between subsequent scans then slice timing introduces some error indeed, but in this case your data is rather problematic anyway. Then your data suffers from motion artefacts and you should probably avoid slice timing completely or at least, replace affected volumes - otherwise you transfer artefacts from an affected volume to neighbouring unaffected volumes]. So for analyses in native space go with ST -> "Realignment: Estimate & Write", when using normalisation go with ST -> "Realignment: Estimate".
If you don't want to start with slice timing, then you HAVE to use "Realignment: Estimate & Write" as far as I can see, even if your preprocessing pipeline includes a normalisation step. Otherwise you correct temporally within your original slices, which is exactly not what you want when starting with realignment. You can try yourself if you "realign manually" via the Display function and flip one volume along y for example (resize(y) = -1). After slice timing the temporally neighbouring volumes should have different values then, depending on whether you use the original data or the time series with the flipped volume. But they do not.
Best,
Helmut
|