Dear Rachel,
I wasn't able to download that picture, and their might be other weird instances leading to results like those you observed, but very likely it's due to normalising the EPI series together with the mean EPI image obtained from realignment vs. normalising without the mean image (or in your case, normalising individually). This is no "error", but expected behavior. During realignment, the mean image is generated based on voxels for which data is available across the whole series. As subjects tend to move over time, and combined with a restricted FoV, it is quite likely that boundary regions are not covered consistently, and corresponding voxels will be coded as 0 (or NaN, dunno right now) in the mean EPI. Now if this image is added in the normalisation batch, corresponding voxels/coordinates will be masked for the whole EPI series. This is a good idea actually, as this way, we can ensure that the smoothed files used for analysis represent the same regions (otherwise, a particular voxel might be non-zero in one image as it was covered in the raw files, but also if it were not covered and "obtained" data at a later stage from voxels nearby during smoothing). If you don't add the mean EPI then the EPI files will be normalised "individually & entirely" as there are no 0 / NaN voxels (assuming the EPI series was only realigned, not realigned & resliced).
> the data are not truncated
I guess it just *looks* as if no files are truncated. It would be necessary to check every single volume, as with the partial coverage your data is prone to even small movements. A jump back and forth in a single volume could have resulted in the observed data loss. Depending on the data one might then simply replace the few bad volumes by neighbouring ones or a mean generated from the remaining "good" realigned ones (via a Realign: Reslice step / reslice mean only, which could then be selected during normalisation instead of the current EPI mean). In case head motion is gradual = region x is no longer covered from vol. y onward it might be necessary to exclude that subject though.
Hope this helps
Helmut
|