Hi all,
to complement Marko's sound comments, there are a few issues to
consider, like slice order acquisition...
I would be tempted to advise:
- For "small" movement, slice timing followed by realignement.
- For "large" movement and ascending/descending acquisition, then
realign followed by slice timing.
- For "large" movement and interleaved acquisition, you're in trouble...
The "large/small" should not be considered as absolute numbers but
rather in term of "displacement per TR", i.e. speed of movement.
For example: a subject slowly shifting by 10mm over a 20min session is
much less problematic, than another one suddenly shifting by 5mm every
now and then.
Do you need slice timing correction at all ?
- if you want to extract the time series at some voxels/ROIs for
analysis outside the GLM framework (ICA, data mining, etc.), yes it
could be useful.
- if it's for a "classic GLM" analysis, no need to do slice timing as
the time difference can be accommodated efficiently by more flexible
basis functions (hrf + derivatives, for example). (Ok this leads to more
complicated models/interpretation, especially for a RFX)
Any way, slice timing only works when the TR is short (<2-3sec), as it
relies on simple linear interpolation !
And a TR of 2-3sec is also precisely the time window when more flexible
basis function for the GLM are working too...
Best,
Chris
Marko Wilke a écrit :
> Hi there,
>
> just my 2 cents:
>
>> 1) Slice timing -> Realign: Estimate & Reslice -> Normalise: Estimate
>> & Write -> Smooth
>
> First, there is a big debate as to whether slice timing is a good idea
> or not (certainly (?) not for block designs), and whether one should
> do it before or after realign. There are people saying that, if there
> is a lot of motion, slice timing cannot do a good job, and thus do
> realign first. Others say they don't want the timecourses to be
> disrupted by realign and thus do slice timing first.
>
> Less controversial: if you normalize anyway there is no need to
> reslice during realign as the parameters will be coded in the headers
> and taken into account during normalization.
>
>> 2) Slice timing -> Realign: Estimate & Reslice -> Coregistration:
>> Estimate -> Segment -> Normalise: Write -> Smooth
>
> Same comments w.r.t. slice timing and reslicing. In my mind,
> segmenting the anatomical is the better option for spatial
> normalization, assuming that you use either spm5 or spm8. The matching
> is much more extensive during segmentation than during spatial
> normalization.
>
>> What is the best order to run the various steps of preprocessing? Do
>> I even need coregistration and segmentation for this?
>
> No in the first case, yes in the last ;) I should also note that
> several people now do the statistics in native space and only
> normalize the con-images for the second level analyses. Save disk
> space, too :)
>
> Best,
> Marko
|