On Mon, 2 May 2005 13:04:34 -0500, Darren Gitelman <d-
[log in to unmask]> wrote:
>Dear Steve / SPM:
>
>Good question. We've recently had the same problem, and have found that
Matthew Brett's software
>tool tsdiffana can highlight trouble in a time series.
Thanks for the tip; I'll have to check it out.
>We have replaced individual volumes with
>averages of the surrounding volumes. We just used 1 volume to each side
but some type of weighted
>average (suggestions?) would probably be better. This somewhat minimizes
the effect of the bad
>volume, but if the effect was due to movement, then often the surrounding
volumes may share some of
>this movement.
>
>Another thing to consider is that with a bad volume, if you do slice
timing correction you may
>spread the badness to other scans.
Right.
>For that reason we first substitute an average, then realign
>(we've also found that Jesper Andersson's unwarping algorithm seems to
also reduce scan to scan
>variance a bit) and then do slice timing.
>
>At the analysis stage what we've done instead is to use user defined
covariates containing ones for
>the bad scan and several surrounding scans. You don't want to convolve
with an HRF since the effect
>of the bad scan is immediate and not delayed. Thus you are modeling the
badness as an effect.
That's a good idea for a practical solution. I'll look into that.
>Perhaps this regressor should also be weighted as an exponential with two
tails but I'm not sure
>what to use.
What's your motivation behind using such weightings? Are you assuming the
artifact somehow "spreads" throughout the time series?
Thanks,
S
>Comments?
>
>Darren
>
>
>==============Original message text===============
>On Mon, 02 May 2005 10:34:22 am CDT "Stephen J. Fromm" wrote:
>
>Is it possible to tweak/augment the SPM code to "censor" a bad fMRI
>timepoint?
>
>"Bad" here means that an isolated timepoint or two has a scanner artifact,
>or excessive motion artifact.
>
>I've thought of two ways to do this:
>(1) Approach taken by AFNI software package: create design matrix
>assuming all data points good, then remove row(s) corresponding to the bad
>observations;
>(2) Replace bad volumes with synthetic data interpolated from nearby
>timepoints.
>
>Any suggestions?
>
>TIA,
>
>S
>===========End of original message text===========
>
>
>
>--------------------------------
>Darren R. Gitelman, M.D.
>Cognitive Neurology and the Alzheimer's Disease Center
>Northwestern Univ., 320 E. Superior St., Searle 11-470, Chicago, IL 60611
>Voice: (312) 908-9023 Fax: (312) 908-8789
>--------------------------------
|