Dear Nicholas,
> While spm_preproc works quite well in most cases, for certain subjects, spm_preproc completely fails to produce a reasonable normalization, and instead yields a heavily distorted brain as the 'normalized' output (when the transform is applied to the anatomical via spm_write_sn). We've had two subjects for whom this is the case; one of which we scanned twice, with spm_preproc producing ridiculous output both times, although in different ways, for a total of three examples of this bizarre behavior of the algorithm.
>
> Please let me know if anyone has had similar experiences, or has any suggestions. Could this be due to the high degree of inclination of the skull in the anatomicals that produce poor outputs?
If the structural scans are fairly far off from the template alignment
(e.g., the high degree of inclination as you pointed out), this can
definitely lead to registration errors (and segmentation errors).
A second common issue is inhomogeneity in the image—this may not be an
issue for you because your other subjects are doing ok.
The general recommendation if there are any issues with segmenting or
normalizing the structural image are to:
1) Reposition the structural image to be in fairly good overall
alignment with the template. This includes getting the origin in
approximately the same place, and making sure the rotations etc. are
not too severe. There are some automated options, but for small
numbers of subjects it's easy to do using the Display button and
adjusting parameters. (Note that if you have co-registered structural
and functional images, you will also need to apply these changes to
the functional images as well…so after repositioning, double-check
that everything is still co-registered.)
2) If need be, running segmentation and saving a bias-corrected
structural image, which you can then segment/normalize again. There
may be more elegant ways of doing this, but the two-pass method seems
fairly robust and easy to implement.
With a combination of 1 and 2, I've never and a structural image fail
to segment and normalize reasonably.
Hope this helps!
Best regards,
Jonathan
--
Jonathan Peelle, PhD
Department of Otolaryngology
Washington University in St. Louis
Office: (314) 362-9044
http://peellelab.org || http://jonathanpeelle.net
|