Hi Alex,
At 05:23 PM 7/1/2005 +0100, Dresner, Alex wrote:
>We're trying to do our fMRI data smoothing in an intelligent way, so
>wanted to assess our FWHM for our data after initial model estimation
>(with no smoothing). But the RPV images have large regions that are
>masked out that are quite important to us. My understanding from Tom
>Nichols' slides (thanks for making those available, Tom!) is that the
>roughness estimate is from the variance of the gradients in the residual
>image. It's obvious from inspection that the volume for which RPV
>measurements are provided is smaller than the Residual volume; is there
>any way to control the gradient or variance estimation process to get
>roughness estimates closer to the 3-D brain surfaces? There's plenty of
>signal in the regions we're interested in, but they are within several
>slices of the lower surface of the brain. Alternately I suppose we could
>just make the measurements in un-related regions of the brain, but I'd
>rather use the relevant areas.
>Maybe this is why people just pick a smoothing kernel that sounds good
>without ever checking the validity of the Gaussian random field theory...
The shrinking RPV image is probably caused by sinc-interpolation during the
smoothness estimation process. In the residual images, voxels outside the
brain are NaN. These NaN values are used in the sinc-interpolation, and
result in NaN RPV values in voxels near the edge of the brain. There is a
newer version of spm_est_smoothness.m function which uses tri-linear
interpolation instead of sinc. See
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind04&L=SPM&D=0&I=-3&P=296111
>And if there are experts out there who care to address this question, let
>me add a simple one onto it: what would be the interpretation of FWHM
>values < 1? Very (spatially) noisy residuals?
From lots of simulation I did years ago, I remember that the FWHM
estimates were usually around 1.2 on unsmoothed Gasussian white noise images.
Hope this helps!
-Satoru
Satoru Hayasaka ==============================================
Post-Doctoral Fellow, MR Unit, UCSF / VA Medical Center
Email: shayasak_at_itsa_dot_ucsf_dot_edu Phone:(415) 221-4810 x4237
Homepage: http://www.umich.edu/~hayasaka
==============================================================
|