I wasn't planning on getting that involved in the single subject ICA stuff, instead just using scan-nulling regressors for dodgy scans (fsl_motion_outliers).
For fsl_glm, does this simply involve cat-ing the null matrix onto the timeseries and running as normal or does a contrast need to be specified telling fsl_glm to output z for the first column only?
Thanks for the help with this!
Jonny
On Wed, Mar 24, 2010 at 12:39 PM, Christian F. Beckmann
<[log in to unmask]> wrote:
> Indeed, you'll need to sort out something mroe snappy as a name, though ;)
> c
Dual Regression Analysis of Spatial Temporal Independent Components = DRASTIC
-Tom
> On 24 Mar 2010, at 10:44, Tom Johnstone wrote:
>
>> On Wed, Mar 24, 2010 at 10:05 AM, Christian F. Beckmann
>> <[log in to unmask]> wrote:
>>> Hi,
>>>
>>> You are right that it is a quasi seed-based functional connectivity type of analysis in the second stage. If the initial spatial maps do contain these nuisance effects then all should be fine, i.e. the output from the first stage should actually give you a full set of time courses, inlcuding time courses that describe the temporal characteristics of the nuisance ffects. Note, however, that certain (subject-unique, mostly) effects might not show up in the group analysis, so yes, adding e.g. motion regressors etc to the 2nd stage is an option.
>>
>> Or performing Melodic de-noising at the individual-subject level
>> ("Dual ICA Dual Regression")?
>>
>> -Tom
>>
>>> hth
>>> Christian
>>>
>>>
>>> On 23 Mar 2010, at 13:16, Jonathan O'Muircheartaigh wrote:
>>>
>>>> Hi
>>>>
>>>> Although ICA itself is relatively immune to outside noise (motion, cardiac etc), is this still an issue at the 2nd Glm stage of dual regression?
>>>>
>>>> My pipeline uses retroicor (the afni implementation) prior to ICA so card/resp noise is as cleanly removed as I can, but at the second Glm stage, is it advisable to add motion regs for each subject as it becomes (in my head) a quasi-functional connectivity analysis on the individual resting datasets?
>>>>
>>>> All the best,
>>>> Jonathan
>>>
>
|