> I've been trying to circumvent my earlier problems with normalisation
> of EPI's by using the procedure described by Christian Büchel
> (https://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0908&L=SPM&P=R895)which
> involves normalising the con* images rather than the EPIs.
It's not a normalisation problem. It's more of an issue of missing
data, and the approach trying its best to fill in the missing values.
> I've been attempting to extend this procedure to beta images, seeing
> as how I am attempting to use time derivative boosted betas in a second
> level conjunction.
>
> My first question is, I suppose, if it is valid to extend the
> framework proposed by Christian to beta images such that I have step
>
> 4a: Normalise boosted beta images to MNI.
>
> 4b: Perform subtraction of beta images after normalisation
You mean contrast images?
>
> 4c: Perform one tailed T-test on subtracted images.
That sounds like a reasonable way to do a second level analysis.
>
> My second question is about the results. Comparing results using the
> procedure proposed by Christian with equivalently smooth data from a
> standard preprocessing pipeline I am left with wildly differing RESEL
> counts. The normal pipeline yields some 1343 resels (2x2x2.75 voxels,
> FWHM 6x6x6) while the DARTEL results yield only 3.1 resels. Some of this
> is, of course, a consequence of the explicit mask applied in step 6 of
> Christian's procedure. However, there seems to be quite a bit of
> resolution loss that's left unaccounted for as FWHM at 6x6x6 with
> 2x2x2.75 mm results in a resel of 66.61 voxels in the DARTELized data.
> However 15692 voxels/66.61 does not equal 3.1. In addition the
> sensitivity for activations seems to have dropped significantly (due to
> the loss of resels, no doubt).
This really does not sound right. I am really not sure why the number
of RESELS should drop to 3.1. I would suggest checking that the images
you enter at the second level are all OK.
>
> Am I doing something obviously wrong somewhere, or is this just how
> the normalise to MNI function works?
It should not work this way. The resel counts etc should remain pretty
much as they were previously (preferably slightly higher, as the
residuals should have relatively less low frequency signal).
Best regards,
-John
--
John Ashburner <[log in to unmask]>
|