Dear Bertram,
In case the first (few) volumes are distorted severely I would just exclude those from preprocessing and adjust the onsets for model estimation. Keep it simple ;)
In case the distortions are that bad that you started to wonder whether they would affect the realignment I would be surprised if they were useful for statistics. Of course, you could still add some dummy regressors to the design matrix, but then again, if the problems occur at the beginning or during the last few volumes of a run it seems to be more straightforward to just not include them instead of including them and then trying to minimize biasing effects on parameter estimation.
Best regards
Helmut
|