Hi John -
In addition to Jeanette's comments, I'm also unsure why you find
handling the multiple runs is so cumbersome - combining them should
simply be a matter of a very simple 2nd level model (you can do a
separate one for each subject) which generates the subject level
contrasts. This can be designed once and copied for each subject.
Cheers,
Eugene
--
Eugene Duff, Phd
Analysis Group, Centre for Functional MRI of the Brain (FMRIB)
Nuffield Department of Clinical Neurosciences
John Radcliffe Hospital
University of Oxford, OX3 9DU
Ph: +44 (0) 1865 222 523
--
On 6 July 2011 20:16, Jeanette Mumford <[log in to unmask]> wrote:
> Hi,
>
> I can think of 2 reasons why you probably shouldn't concatenate, even after
> preprocessing. First, the autocorrelation of concatenated time series will
> have a weird behavior at the concatenation points, which can cause issue
> with the prewhitening. More importantly, if you have a single run that has
> a higher variability (maybe there were scanner issues or the subject wasn't
> paying as much attention or they moved more, etc) in the concatenated
> analysis you are effectively penalizing all of your runs instead of only the
> bad run. If the runs are analyzed separately and then combined, the "bad"
> run will be down-weighted by its higher variance and the "good" runs will
> have less of a penalty based on their smaller variances.
>
> There are probably other reasons as well. I'm unsure if the grand mean
> scaling would also be negatively impacted by this, since typically that is
> applied on a run-by-run basis and is a necessary step to make data
> comparable across runs and subjects.
>
>
> Cheers,
> Jeanette
>
>
>
> On Wed, Jul 6, 2011 at 1:17 PM, John Herrington <[log in to unmask]> wrote:
>>
>> I have an experimental design with multiple between and within-subjects
>> factors, split across multiple experimental runs (by runs, I mean separate
>> fMRI time series acquisitions, separated by a minute or so during which no
>> MRI data are collected). The multiple runs share a common baseline
>> condition, but vary in the attentional load for the experimental condition.
>> It would make the analysis of these data much less cumbersome to
>> concatenate the fMRI data together into a single run and submit this
>> concatenated file to GLM. This practice has been discouraged multiple times
>> on this listserv, but I have not been able to determine from these posts if
>> concatenation is *always* a problem, or if it is a problem only in some
>> instances – namely, before pre-stats have been carried out (motion
>> correction and temporal filtering). My questions:
>>
>> 1) If one were to run pre-stats on each fMRI time series, then register
>> them together (either directly or by placing all of them in MNI space),
>> would it then be acceptable to concatenate them for submission to GLM?
>>
>> 2) If not, why not? What problems would remain that were not handled by
>> running pre-stats separately on each time series?
>>
>> 3) Are there any additional preprocessing steps that could fix these
>> problems?
>>
>> Thanks,
>>
>> John
>
>
|
