VBM data are p values, constrained to range from zero to one. Close to
zero, as will be the case outside the brain, your data are strongly
skewed, and also very likely quite kurtotic. Thus, this 'noise' does
not behave the way you are used to in fMRI data. Whatever you do with
repeated segmentation steps, you will not obtain VBM data with a
distribution behaving properly at very low values.
All the best,
Quoting "Woicik, Patricia" <[log in to unmask]>:
>> We are using DARTEL (following the DARTEL manual)for our VBM
>> studies and while our exploratory analysis (very low threshold) we
>> see some 'activations' outside the brain. These activations are not
>> significant but they are there. If there is indeed something
>> there, then we can not trust our results and there is definitely
>> something that went wrong during the pre-processing. We also seem
>> to think that these activations outside the brain are just noise,
>> and since we are looking at very low thresholds we are just
>> picking up noise. But this doesn't happen when we perform
>> exploratory fMRI analysis.
>> If this is indeed due to inadequate spatial pre-processing then
>> should we segment twice to get rid of whatever is left from the
>> previous segmentation step. I propose the following pre-processing
>> 1. Spatial Normalization
>> 2. Segmentation
>> 3. Averaging all GM and WM images and using it as a prior for a
>> second segmentation on previously segmented images.
>> 4. Use DARTEL tools.
>> Has anyone tried something like this before?
>> Many thanks for your help.
>> Patricia A. Woicik, Ph.D.
>> Neuropsychoimaging Group
>> Brookhaven National Laboratory
>> Bldg 490, Upton, NY 11973-5000
>> (631) 344-4472