Hi Gemma
Are you perhaps referring to the issue of nonstationarity in VBM data
(mentioned in Ashburner & Friston, 2000)? The issue isn't one of the
smoothing kernel (which is typically isotropic), but rather that the
data themselves are not equally smooth over space. The Gaussian
smoothing is not enough to overcome the inherent nonstationarity in
the data.
One consequence of this is that by chance, larger clusters tend to
occur in regions where the image is more smooth, and smaller clusters
where the image is less smooth. This invalidates the typical random
field-based cluster-level correction in SPM. However, the
nonstationarity toolbox will correct for this:
http://fmri.wfubmc.edu/cms/NS-General
Hope this helps,
Jonathan
On Fri, Mar 27, 2009 at 10:16 PM, Gemma Monte <[log in to unmask]> wrote:
>
> There is a concept I am confused about: (in pre-processing of VBM) I thought
> that the convolution with the Gaussian over the data is describing the
> smoothness in direction k. This leads to the data aren't non-uniformly
> (non-isotropic) smoothed. But when we apply the smooth to the data,
> we introduce the value in the x, y and z directions, why is non-isotropic?
>
> And if the data are non-uniformly smoothed, it has an influence over the
> results, which are these effects over the results? How it is possible to
> correct for them?
>
> Best regards,
> Gemma
>
>
>
|