Hi Antonios-Konstantinos,
I've just looked at your results, and they're perfectly fine and completely makes sense... Be sure to threshold the clustere/clusterm map at 0.95, and for the tfce, either 0.95 or 0.99.
One small thing though: there are still some extra-brain voxels after bet or most of your subjects, so you might want to erode your mask for randomise a bit or threshold it a bit higher (originally created by taking the mean of GM_merg across time and thresholding the mean at 0.01).
Cheers,
Gwenaelle
--- En date de : Jeu 7.5.09, Antonios-Konstantinos Thanellas <[log in to unmask]> a écrit :
> De: Antonios-Konstantinos Thanellas <[log in to unmask]>
> Objet: [FSL] VBM analysis went wrong
> À: [log in to unmask]
> Date: Jeudi 7 Mai 2009, 15h26
> Dear fsl users,
>
> I have a dataset consisting of 2 groups of subjects (31
> Controls- 17
> Alzheimer's patients). The Alzheimer's dataset
> consists of T1 scans that
> were taken 6 months after the "onset" of the
> disease. I did VBM analysis
> using a kernel of 3 cause I was expecting small to
> intermediate changes
> (since my scans depict an early stage of the disease) and
> after
> fslvbm_3_proc, 4 volumes out of 48 were distorted. I
> excluded these 4 vols
> from the GM_mod_merg_s3 files and continued with randomise
> choosing a
> threshold of 3 for cluster size and cluster mass tests.
> Though my results
> (tfce_corrp_tstat,clustere_corrp_tstat and
> clusterm_corrp_tstat) were really
> strange revealing huge clusters covering 60-70% of the
> whole brain with very
> high p-values < 0.003. It's not logical to expect
> such severe changes and I
> cannot point out what went wrong in the analysis.
> I would really appreciate your help.
> Antonios-Konstantinos Thanellas
|