Dear Rachel,
I was trying to find an explanation why one could run into issues like those you've observed, with some assumptions that actually don't hold ;) So forget everything what I just wrote except the general statements on head motion as a cause for data loss, sorry for that.
Starting with the simple part, the EPI mean. Registering on the mean is a two-stage procedure, in the first step the EPI images are realigned on the first image of the series, based on which a mean image is created internally, and in the second step the images are realigned onto that mean image. Thus, the voxel space (including the orientation of the "voxel grid") of the mean image corresponds to that of the first (raw) image from the series. It would be necessary to look into the code for the details, but at that stage, the mean EPI does *not* reflect "voxels covered consistently" but something like the average of the series (possibly there are some restrictions like having data for a certain percentage of volumes above a certain intensity). Now, even if you reslice everything during the realignment (which we usually don't want to do to avoid an interpolation step), then only the resliced EPI images are masked consistently, but not the written-out EPI mean (in contrast to what I had said previously). In other words, when looking at the EPI mean you don't get a good impression of the brain coverage across the whole series.
Thus, normalized, realigned EPI files can be truncated to a different extent, depending on brain coverage in the different files. The normalized EPI mean might look alright = as if you had covered a certain region entirely, while a normalized image from the series actually does not - just as if you compared the unnormalized EPI mean with the corresponding unnormalized image from the series.
With regard to normalization differences in coverage depending on forwarded images, I observed the following:
1) When forwarding realigned f* files only the different normalized volumes are truncated to a different extent.
2) When forwarding realigned f* files + EPI mean the different normalized volumes are truncated to a different extent; the truncations for the normalized EPI series are not identical to those obtained with 1).
3) It does not make a difference for the EPI mean whether it is normalized on its own vs. together with the EPI series.
Which does not really agree with your observations though.
But leaving this aside for the moment, your structure seems to get truncated as well?
Best
Helmut
|