Hi,
Your experiment sounds perfectly suited to the three-level approach
described in the documentation. It will combine the GLM outputs
from each level at each stage, so the second-level combines your
first-level parameter estimates and variance estimates, which is
preferable to concatenating the runs and will not harm your
ability to detect activation even though your first-level results
may look poor. If you read the documentation and the FSL course
slides then this should explain how things work.
All the best,
Mark
On 13 Nov 2008, at 14:03, Yael Shani wrote:
> Hi
> I'll try to be more clear. I have a session with five runs. The
> session contains 80 trials (Each run contains 16 trials).
>
> I don't understand how the second level analysis works. When you
> say it merges all the first level analysis runs, do you mean it
> takes every contrast I made and average the results from every
> first level analysis? if so, since my runs contain very few trials,
> I don't suppose I will see anything significant.
>
> The documentation in the site relate to a combination of second and
> third level analysis, which I'm not there yet :)
> I'll be happy if you explain how the second level analysis works
> Thank you again
> Yael
>
> On Thu, Nov 13, 2008 at 10:33 AM, Mark Jenkinson
> <[log in to unmask]> wrote:
> Hi,
>
> Just to add that we really do discourage the merging of timeseries
> data as it
> messes up the pre-processing - especially the temporal filtering.
> You are
> much better adding a separate level where you average over the runs
> using
> the GLM, with the first level just doing each run separately. This
> should be
> fine if you have 5 runs per subject. Have a look at the
> documentation at:
> http://www.fmrib.ox.ac.uk/fsl/feat5/
> detail.html#MultiSessionMultiSubject
> for more details.
>
> All the best,
> Mark
>
>
>
> On 12 Nov 2008, at 20:19, Eugene Duff wrote:
>
> Hi Yael,
>
> It is a bit unclear what you need to do. If you just want to do an
> typical single-subject multi-session analysis you probably do not
> need to manually merge your data. You run an ordinary first-level
> analysis on each session, then run a higher-level analysis in Feat,
> where you include each of the 1st level Feat directories as
> inputs. Feat then automatically merges the appropriate 1st level
> datafiles before it does the modelling. Otherwise, fslmerge is the
> tool to merge data.
> Eugene
>
>
> 2008/11/12 Yael Shani <[log in to unmask]>
> Hi Reza
> I'm not sure I understand. When you say merge do you mean the in
> the second level analysis feat takes all the runs and concatenate
> them into one long run or averages the results of all the runs?
> Thanks again
> Yael
>
>
> On Sun, Nov 9, 2008 at 5:45 PM, Reza Salimi <[log in to unmask]>
> wrote:
> Hi Yael,
> to merge some volumes, you can use fslmerge which does it in
> different directions(x,y,z,t)
> and also, in case of using a typical feat analysis, this merging
> happens itself as a part of the group-level analysis and
> merged result is saved as filtered_func_data.nii.gz
> hope it helps.
> cheers
>
>
> On Sun, Nov 9, 2008 at 3:39 PM, Yael Shani <[log in to unmask]> wrote:
> Hi FSL'rs
>
> I want to do a within subject analysis, but before I need to
> understand how to merge data.
> Each subject had 5 runs. I would like to treat these runs as one
> long session. How do I do that? As far as I understand the second
> level analysis with fixed effects does not concatenate. Am I right?
> Many thanks
> Yael
>
>
>
> --
> G. Salimi-Khorshidi,
> D.Phil. Student, Dept. of Clinical Neurology, University of Oxford.
> [log in to unmask] http://www.fmrib.ox.ac.uk/~reza
> FMRIB Centre, John Radcliffe Hospital,
> Headington, Oxford, OX3 9DU
> Tel: +44 (0) 1865 222466 Fax: +44 (0)1865 222717
>
>
>
>
|