 |
 |
I have a question regarding when to apply MELODIC in the post below. I ran a first level analysis with motion parameters added for all subjects. In a separate analysis, I ran the same first level analysis, except with Melodic turned on and without adding motion parameters to the model. Adding motion parameters to the model cleaned up the data, but in some of the runs head motion appears to be in a single TR (that is tempting to try to clean up using melodic ICA). Can I use Melodic to remove head-motion artifacts - and then rerun the analysis with the motion parameters added? I don't want to run smoothing twice. (and I can't select add motion parameters to model with prestats turned off). Is there a valid way to do this, or is it only valid to run the analysis with Melodic or with the motion parameters added to the model - and not both? Thanks for the clarification. On Thu, 3 Apr 2008 21:23:31 +0100, Christian F. Beckmann <[log in to unmask]> wrote: >Hi > >On 3 Apr 2008, at 16:54, Carlos Faraco wrote: >> Steve, >> >> Thank you for your reply. Just have a few more questions though. >> >> 1) In the FSL manual it says to use MELODIC before the data is >> processed >> (and BETed I imagine). I guess this is supposed to make it easier to >> detect >> motion if you see "activity" closer to the skull. My question then >> is, does >> it matter when MELODIC is applied? > >This also is an area where one can argue both ways - if you want to >use artefact related to the MR physics (slice dropouts, say) then in >most cases you want to be as close to the original data as possible >and use the non pre-processed data without betting/mc/smoothing. If - >like in the case of head motion - you want to correct for artefact >over and above what you can achieve with other approaches (i.e. rigid >body motion correction) then clearly you'd want to perform these steps >first. > >> >> >> 2) Are there any pictorial representations of what motion in the >> MELODIC >> output would look like? As I stated previously, I feel very hesitant >> to just >> go in there and start judging things as motion without any type of >> criteria. >> > >these typically look like rim-like artefacts - see the melodic course >talk for some examples (http://www.fmrib.ox.ac.uk/fslcourse/lectures/melodic.pdf > ) > > >> 3)If motion parameters are added to the GLM, does this have to be >> done for >> all subjects (since the degrees of freedom would be reduced)? Would >> the same >> be true for MELODIC noise removal, even though in that case the >> number of >> reduced DFs might vary per subject? >> > >Differences in the DOFs between subjects are accounted for in the >higher level analysis of FEAT, i.e. adding motion regressors for some >but not all subjects is OK. When using melodic denoising then this >more difficult, though you can adjust the dofs explicitly on the >command line when using ttoz for first level analysis - for the higher >level analysis you'd want to adjust the dof files using fslmaths >before running higher-level analysis. > >hth >Christian > > >> Thank you, >> >> Carlos Faraco >=========================================================================