Dear Timo,
>We're wondering what is the best way to rule out patients who have moved
>too much during visual fmri study. The way we have used to do outruling
>patients, although with FSL,
>
FS what?
> is by looking the histogram of first
>realigment and excluding patients with overall movement of 1 mm.
>Instead of this our collegues proposed for us to run the realigment again,
>for the motion corrected data, to see, wether or not the realigment on the
>first time was succesfull. Again by looking the histograms, this time from
>second realigment, we decided to exclude patients with residual movement
>of 0.1 mm. The idea was that if the first realigment was not succesfull
>there should be "movement" in the second histogram and vice versa.
>
>>From 62 patients around 20 was excluded with the method number one, but
>using the latter resulted in no-one to be excluded. Some of those
>previously excluded ended up having completely irrational activations.
>
>
>
There are a number of reasons why realignment of fMRI data do not do a
perfect job (and with perfect I mean "recreating that data that would
have obtained if there had been no movement"). These are all caused by
the inadequacy of the rigid-body model to correct for all the effects
that movement cause. I won't go into details, but things to watch out
for are i) Sudden movements, ii) Rotations around the x- and or y-axis
and iii) large through-plane movements.
From this perspective it makes some sense to use a simple heuristic
like "through everything away with movement larger than some arbitrary
limit". I suspect your suggestion of 1mm is as good as anyones.
The second "test", i.e. looking for estimated movement parameters in a
second realignment does NOT make sense. That simply tests how well the
realignment procedure was implemented in the first place. If all details
are handled correctly and the same interpolation model was used in the
estimation of the movement parameters as in the actual reslicing there
should be "no" residual movement. N.B. that this is VERY different from
residual movement related variance (of which there is surely plenty).
Good luck Jesper
|