Hi Pablo
> I am currently working on a study of patients who went through surgery and had removed regions of their brain. SPM8 does a good job on segmenting and normalizing all the patients but one, who has a big resection. In this patient segmention is not done properly and I cannot find out why.SPM does agreat job with other patients with a resection as big as this one´s.
> żIs there another way of normalizing for particular cases like this?
The problem is likely the large mismatch between the expected tissue
classes (i.e., the tissue probability maps SPM uses to aid in
segmentation, which are based on normal brains) and the actual anatomy
of your patient's brain. I.e., the priors are in this case providing
wrong information because the patient's brain is not normal.
I haven't tried this yet myself, but you may want to look at this
paper from Seghier et al. in which they included an extra tissue class
to help model lesioned areas:
Seghier et al. (2008) Lesion identification using unified
segmentation-normalisation models and fuzzy clustering. NeuroImage 41,
1253-1266.
http://dx.doi.org/10.1016/j.neuroimage.2008.03.028
Even if you don't use the same "lesion identification" process, I
wonder if being able to model the lesion as its own tissue class would
give you a more accurate segmentation. Maybe someone else on the list
has tried this?
Good luck!
Jonathan
|