| Is it known what is the optimal amount (if any) for spatial filtering of
| brain images for accuracy in
| a) spm99's within-modality rigid body registration (realignment)
| b) spm99's atlas matching (normalization)?
|
| For comparison, AIR's rigid body PET-PET registration does better with
| fairly heavy smoothing.
I haven't done any extensive testing with different amounts of smoothing.
For the record, spatial normalisation uses 8mm FWHM smoothing, fMRI
realignment uses 6mm FWHM, PET realignment uses 8mm FWHM and coregistration
via the Coreg button uses 8mm.
An optimal filter can probably be determined using Wiener filter theory,
based on the power spectrum of the true signal, and the power spectrum
of the noise. I haven't properly thought about this, but suspect that
the principles should be applied to the gradients of a mean image to attempt
to find the "true" signal, and the gradients of some difference images
to estimate the noise. Similar principles could probably also be applied
when deciding on the optimal smoothing to use before running the stats.
Best regards,
-John
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|