Hi - this sounds ok, but your method of generating data for approach 2, by
splitting up the results of approach 1, is probably not optimal - in
practice you'll still probably have some effects in the temproal
filtering, and varying amounts of smoothness induced by the unusual way of
doing motion correction. better to compare the approaches, if that's what
you want to do, byt doing each totally separately!
Cheers.
On Mon, 13 Sep 2004, Xun Liu wrote:
> Just to follow up this thread, I finished the analysis of Approach 2 below.
> Besides the conceptual complexity, Approach 2 also took much much longer
> time (e.g., one cope#.feat took more than 10 hours with 12 subjects).
> However, the results are not much different from those of Approach 1 (with
> concatenated runs). There are some trivial differences but the major
> clusters/patterns stay the same, especially after thresholding.
> One thing should be noted though. To better compare the two approaches, I
> took the pre-processed time series from Approach 1 (i.e.
> filtered_func_data -- this is the data after all pre-processing steps but
> before FILM prewhitening, right?) and split them into separate runs, in
> order to keep the pre-processing part constant. Though there were some
> subjects with larger across-run motion, I took care to motion correct those
> with two-stage motion correction (within-run and then across-run) already
> when I went with Approach 1. The spatial filter and temporal filter (high-
> pass) should not be affected either. In terms of the signal intensity
> differences across runs, I had two constant EVs to model the mean shifts of
> the second and third runs with regard to the first run in Approach 1.
>
> P.S. When it is doing the estimation, some times it pops up the "ndtri
> domain error" message when the percentage numbers progress. But I think the
> results are not affected by this. Any idea of what the message means?
>
>
> On Thu, 9 Sep 2004 16:21:05 +0100, Stephen Smith <[log in to unmask]>
> wrote:
>
> >Hi,
> >
> >On Mon, 6 Sep 2004, X Liu wrote:
> >
> >> Thanks very much. I thought the original question is going to be buried
> >> without being answered. I've done Approach 1 and am holding to re-do it
> >> with Approach 2 until I get some expert opinion on it. The second
> approach
> >> definitely requires more labor. Are there any strong reasons against the
> >> first approach (e.g. statistical validity, filter efficiency, motion
> >> correction across runs, etc.)? I will go ahead and anlyze the data with
> >> Approach 2 and see how much the end results differ.
> >
> >Yes - there's various comments in the FSL email archive on this. Time
> >series concatenation is nasty as the 1D signal doesn't "join up" well
> >across runs, with nasty analysis implications....
> >
> >> P.S. Another message (Distributed Computing Once More) I sent on Aug. 30
> >> did not get people involved to discuss. Hopefully FSL development team
> and
> >> other experts on this issue can look into it to see if it is useful. One
> of
> >> the advantages of FSL is that it enables easy batch processing (mostly
> >> serially) with scripts. Batch parallel processing has to be able to run
> on
> >> distributed/clustered system to gain full power.
> >
> >Yes, in the forthcoming release, the diffusion toolkit will use parallel
> >processing - for other areas like multi-session FEAT analysis or FLAME
> >runs, UCLA are doing some really nice work on this, so watch this
> >space....
> >
> >Thanks, Steve.
> >
> >>
> >> On Sat, 4 Sep 2004 09:21:50 +0100, Stephen Smith <[log in to unmask]>
> >> wrote:
> >>
> >> >Hi, yes that all makes sense. Approach 2 is what you want and you're
> right
> >> >to take out the constant-EV ev1 from your original email.
> >> >
> >> >Cheers.
> >> >
> >> >
> >> >On Tue, 31 Aug 2004, Xun Liu wrote:
> >> >
> >> >> On Tue, 31 Aug 2004 21:17:33 +0100, X Liu <[log in to unmask]>
> >> wrote:
> >> >>
> >> >> >Concatenate run/session or not?
> >> >> >
> >> >> >We have a design with 2 independent variables (IV1 and IV2), each
> with 3
> >> >> >levels. IV1 is manipulated across blocks within a run and IV2 is
> >> >> >manipulated across runs. There are two approaches I can think of to
> >> anaylze
> >> >> >the data for the main effects and interaction for this design. What
> is
> >> the
> >> >> >advantage and disadvantage of each approach, from the conceptual and
> >> >> >practical perspectives? Is one more valid than the other from the
> >> >> >statistics point of view? Thanks very much.
> >> >> >
> >> >> >
> >> >> >Approach 2:
> >> >> >Analyze each run separately and model just the 3 conditions/levels of
> >> IV1
> >> >> >(ev1, ev2, ev3). Then the main effects of IV1 can be set up as below.
> >> >> >
> >> >> >Contrast ev1 ev2 ev3
> >> >> >mean (1) 1 1 1
> >> >> >IV1 (2) 1 -1 0
> >> >> > (3) 1 0 -1
> >> >> > (4) 0 1 -1
> >> >> >
> >> >> >And then proceed to the group analysis with the EVs/contrasts setup
> as
> >> >> >below for the three levels of IV2 (say for 5 subjects)
> >> >> >
> >> >> >Then the .gfeat folder should include 4 cope#.feat subfolders, one
> for
> >> each
> >> >> >of the contrasts from the first level. zstat2 to zstat4 of cope1.feat
> >> will
> >> >> >assess the main effect of IV2 (zstat1 is the overall grand mean of
> both
> >> IVs
> >> >> >again baseline). zstat1 of cope2.feat to cope4.feat will assess the
> main
> >> >> >effect of IV1. zstat2 to zstat4 of cope2.feat to cope4.feat will
> assess
> >> the
> >> >> >interactions.
> >> >>
> >> >> Just found out an error. To correct myself, ev1 is a linear
> combination
> >> of
> >> >> ev4 to ev8. So take out ev1 and contrast 1 is just the combination of
> the
> >> >> new ev3 to ev7 (weighted or not WRT to the other contrasts?). The
> >> >> interpretation of the main effects and interaction above stands.
> >> >>
> >> >> Group ev1 ev2 ev3 ev4 ev5 ev6 ev7
> >> >> 1 1 1 1 0 0 0 0
> >> >> 1 1 1 0 1 0 0 0
> >> >> 1 1 1 0 0 1 0 0
> >> >> 1 1 1 0 0 0 1 0
> >> >> 1 1 1 0 0 0 0 1
> >> >> 1 -1 0 1 0 0 0 0
> >> >> 1 -1 0 0 1 0 0 0
> >> >> 1 -1 0 0 0 1 0 0
> >> >> 1 -1 0 0 0 0 1 0
> >> >> 1 -1 0 0 0 0 0 1
> >> >> 1 0 -1 1 0 0 0 0
> >> >> 1 0 -1 0 1 0 0 0
> >> >> 1 0 -1 0 0 1 0 0
> >> >> 1 0 -1 0 0 0 1 0
> >> >> 1 0 -1 0 0 0 0 1
> >> >>
> >> >> Contrast ev1 ev2 ev3 ev4 ev5 ev6 ev7
> >> >> c1 0 0 0.2 0.2 0.2 0.2 0.2
> >> >> c2 1 0 0 0 0 0 0
> >> >> c3 0 1 0 0 0 0 0
> >> >> c4 1 -1 0 0 0 0 0
> >> >>
> >> >
> >> > Stephen M. Smith DPhil
> >> > Associate Director, FMRIB and Analysis Research Coordinator
> >> >
> >> > Oxford University Centre for Functional MRI of the Brain
> >> > John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
> >> > +44 (0) 1865 222726 (fax 222717)
> >> >
> >> > [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
> >>
> >
> > Stephen M. Smith DPhil
> > Associate Director, FMRIB and Analysis Research Coordinator
> >
> > Oxford University Centre for Functional MRI of the Brain
> > John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
> > +44 (0) 1865 222726 (fax 222717)
> >
> > [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
>
--
Stephen M. Smith DPhil
Associate Director, FMRIB and Analysis Research Coordinator
Oxford University Centre for Functional MRI of the Brain
John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
|