At 01:29 PM 11/9/2005 +0000, Stephen J. Fromm wrote:
>I was wondering if anyone would care to comment on a method for dealing
>with isolated bad volumes in data. An example would be a volume that's
>rendered unusable because of movement during the volume. (Leads to
>alternating slice brightness.)
>
>One cannot merely discard the entire timepoint, because that would lead to
>a discontinuity in the time series.
>
>For the statistics itself, I add a "user-specified" regressor singling out
>that individual timepoint, to "regress out" the effects of that volume.
>
>At the preprocessing stage, the volume must be replaced if slice timing
>correction is used, else the data corruption will "spread" in time.
>Currently I merely replace the volume with the average of the neighboring
>volumes.
Dear Stephen,
A Simple way with dealing with bad volumes is to change the weighting
matrix in your SPM (W) and set the weight of these volumes to something
very low. I think this is the most elegant way to do this.
A toolbox that does this exclusion automatically for you is the RobustWLS
toolbox for SPM.
http://www.bme.jhu.edu/~jdiedric/imaging/SPMj.html
If you just throw out the volumes and re-concatenating your volume to a new
time series, you have to be careful that you get into trouble with the
assumptions of temporal dynamics (i.e you have to throw that frame out of
you design matrix and out of your temporal variance-covariance matrix as well.
best,
Joern
|