Hi,

I'd it is in fact not so much the pure amount of motion. Instead, stimmulus-correlated motion can introduce false-postive activations (or, when you include the standard or extended motion parameters into your model, may also decrease the sensitivity of your GLM to detect the true positives).
On the other hand, stimulus-uncorrelated motion tends to increase the false-negative results.
Even subvoxel motion can be a disaster while a single huge motion spike in the times series may not be a problem at all.
Cheers,
Andreas

Von: Jesper Andersson <[log in to unmask]>
Antworten an: FSL - FMRIB's Software Library <[log in to unmask]>
Datum: Mittwoch, 14. August 2013 16:38
An: <[log in to unmask]>
Betreff: Re: [FSL] Head motion limit ?

Dear Ramy and Antoine,

I think we should be a little careful with "hard and fast" rules for "too much motion". We can all agree that less motion is better, but I would hesitate to advice someone to throw data out based on a rule like this.

One thing is that different types of movement result in different residual effects after motion correction. A (close to) pure translation is easier to correct than a rotation around the x- or y- axis as the latter will cause changes to the field that will change the pattern of distortions and drop-out, something that isn't dealt with in most registration algorithms.

It is also the case that a slow gradual movement is easier to correct than a sudden one (though in the latter case one can often fix it through a stick-regressor in the subsequent modelling).

Finally there is the important distinction between movement which are unrelated to the task and that which is caused by/correlated with the task, where the latter is a much bigger problem than the former.

So, Antoine, where I am concerned there is no easy rule and you will have to use your best judgement in each study.

Jesper


On 14 Aug 2013, at 15:08, ramy kirollos <[log in to unmask]>
 wrote:

Hi Antoine, 

Generally, the rule of thumb is that you don't want the head motion to exceed half a voxel size.  So if you selected voxel sizes of 1mmx1mmx1mm, motion should not be greater than .5mm.  

Cheers,

Ramy


On Wed, Aug 14, 2013 at 9:29 AM, Antoine BERNAS <[log in to unmask]> wrote:
Hi everybody,

After preprocessing on fMRI datas, I have some subjects with more or less big head motions (looking at the outputs of MCFLIRT).
Does anyone know if there is a limit in the relative or absolute mean head motion, beyond which it would be better to exclude the subject ?

Best regard,

Antoine B.