|     I believe that the SPM99 realignment is robust with respect to
| correcting for large movements (e.g., between run repositioning by the
| subjects) however, it has been our experience that this can sometimes be
| substantial (greater than 10 mm or degrees, especially with patient
| populations.

It is true that SPM99 is more robust to larger displacements.  However,
I would probably recommend being slightly more cautious of data where large
rotations are involved (don't ask me how large, because I don't have a
good rule of thumb for this yet).  Rotations introduce relative distortions
into the data that can not be corrected by simple rigid-body alignment.
This is something that the methods group at the FIL is currently looking into
solving.

|             Since we analyze the runs as separate sessions in statistics,
| I prefer to realign them separately.   Also, we (and others) have noticed
| that there can be substantial differences in the global mean (across all
| voxels within run) between sessions.  For example, the image intensity for
| one session could range from 6000-8000 (mean of 7500), while another run may
| range from 8000-10,000 (mean of 9500).    We can control this somewhat in
| the reconstruction of our images, but these 'gain' differences do exist.
| SPM realignment may be robust to this change in means across sessions, but I
| preferred not to test this.

SPM should be robust to differences in global mean, as there is a parameter
in the registration model that accounts for a uniform intensity scaling
difference between the images.

|                                    We typically then realign within session,
| create mean images, then normalize these mean images to the EPI template and
| then apply these parameters to the rest of the images within the respective
| session.

When realigning multiple sessions of the same subject, all images within
session are aligned with the first image in the session, and all the first
images of each session are aligned with the first image of the first session.
The realignment parameters are then combined.  Assuming that there are
large systematic differences between sessions, then this should result in
better within session alignment, but also mean that images from all sessions
are aligned together.

If the "adjustment" is used for writing the realigned images, then this is
applied a session at a time.

|            Another tip I have implemented here is to reposition (using the
| new display facility in spm99) all of the images within each run so that
| they closely resemble the position of the EPI template.  This really appears
| to make the spatial normalization more robust.

Sound advice.  This should be the first thing to try if a spatial
normalisation has gone badly wrong.

Cheers,
-John



%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%