The attached images don't seem to have been corrected for multiple
comparisons, in which case you can't say that there are areas with
significantly lower FDG uptake. With an uncorrected threshold of
0.001, I would expect to see a few blobs appearing by chance.
Therefore, the difference could be explained by noise (although I
suspect that you'd see stronger differences if you do a paired t test
to compare the data pre-processed in different ways).
Be careful when you manually edit the rc1 images. There are a couple
of matrices in the headers that you should be careful about not
changing.
Best regards,
-John
On 9 September 2012 12:38, Emiko Morimoto <[log in to unmask]> wrote:
> Dear John and SPMers,
>
> I'd like to compare PET images between each patient and normal volunteers by SPM8.
> The PET images were coregistered to MPRAGE images and were normalized to MNI space through the DARTEL template.
> I found many areas of significantly lower uptake of FDG in the surface of the frontal lobe in each patient (see the attached figure).
> They seemed to be not in the parenchymas, so I thought they were apppered due to fitting error.
> The frontal lobes of the patients images normalized to MNI space are seemed slightly smaller than that of normal volunteers normalized to MNI space.
> I suspected one of the reasons was segmentation error by New Segment, because parts of the skull and the SSS were remained in rc1 images (see the attached figure) and the areas of significantly lower uptake in the surfaces of the frontal lobes were reduced (not all) after deleting missegmented areas in a patient's rc1 image by hand.
> How can I get clean rc1 images? (It takes too much time to delete dust by hand !) What do you think of other reasons for the surface errors? Please tell me the good way to reduce errors in the statistical results.
>
> Sincerely
> Emiko
>
|