spm does not orthogonalize the columns of the design matrix EXCEPT it does
orthogonalize basis sets. So, for example, the temporal derivative is
orthogonalized wrt the canonical hrf. However, column 2 is not
orthogonalized wrt column 1, etc. When you initially make the design matrix
the block columns (sessions) look all white. After you estimate it they look
like pic1. This is due to the incorporation of filtering and sphericity
effects in the design. Anyway it has the effect of removing the session
means. (It's actually not quite right as the block effects should be in the
confounds part of the design (xX.iB), but the result is quite similar.)
darren
----------
> -----Original Message-----
> From: SPM (Statistical Parametric Mapping)
> [mailto:[log in to unmask]] On Behalf Of <Laurence> <Wang>
> Sent: Friday, January 11, 2008 5:28 PM
> To: [log in to unmask]
> Subject: Re: [SPM] regressors for concatenation
>
> Hi Darren,
>
> Thank you for your prompt reply! But I am still a little bit
> puzzeling on this issue. if the session- mean regressors are
> treated as covariete regressors (attached pic1), they will be
> orhogonalised to the condition regressors, in this case, it
> seems the effect of different session-mean can NOT be removed
> out. we also tried to treat them as condition regressors
> (attached pic2), but I am not sure if it is feasible becaue
> they will be convolved with HRF, which you recommended not to
> do in such a way. Could you please have a close look at the
> attached pictures and explain more.
>
> Thank you so much!
>
> Laurence
>
> On Fri, 11 Jan 2008 00:27:18 -0600, d gitelman
> <[log in to unmask]> wrote:
>
> >i meant hrf.
> >
> >covariate. you don't want them convolved with the hrf.
> >
> > darren
> >----------
> >
> >> -----Original Message-----
> >> From: <Laurence> <Wang> [mailto:[log in to unmask]]
> >> Sent: Thursday, January 10, 2008 11:10 PM
> >> To: [log in to unmask]; Darren G
> >> Cc: <Laurence> <Wang>
> >> Subject: regressors for concatenation
> >>
> >> Hi Darren and SPMers,
> >>
> >> In order to eliminate the effect of different session means when
> >> concatenate the multi-session data, we created the block-type
> >> regressors.
> >> My questioin is that these regressors should be entered into the
> >> design matrix either as "condition" (in this case, they will be
> >> convolved in the estimation ) or as covariate regressors, as the
> >> attached picture showed (ses2-ses6)(in this case,they will be
> >> orthogonal to the conditions). Could you please tell me
> which option
> >> is appropriate during processing the concatenated data?
> >>
> >> Thanks in advance
> >>
> >> Lawrence
> >>
> >> On Sun, 17 Jun 2007 18:36:57 -0500, d gitelman <d-
> >> [log in to unmask]> wrote:
> >>
> >> >Luke
> >> >
> >> >> -----Original Message-----
> >> >> From: SPM (Statistical Parametric Mapping)
> >> >> [mailto:[log in to unmask]] On Behalf Of Luke Stoeckel
> >> >> Sent: Sunday, June 17, 2007 12:26 PM
> >> >> To: [log in to unmask]
> >> >> Subject: [SPM] DCM with multiple sessions per subject
> >> >>
> >> >> For some reason, the original message was not included
> >> with my reply.
> >> >> Please see the issue below. Thanks.
> >> >>
> >> >> DCM mavens:
> >> >>
> >> >> We have collected 6 runs (not repetitions) of block-design
> >> fMRI data
> >> >> for each subject. I want to test a model using DCM
> >> including the data
> >> >> from all 6 runs. In a PPI analysis, this was simple...I
> >> would extract
> >> >> the time series for a given VOI (using the same seed voxel
> >> and sphere
> >> >> dimensions) for each run separately and create a model
> including 6
> >> >> sessions.
> >> >> However, it does not appear to be that simple using DCM
> in SPM5. I
> >> >> have read through the postings about this issue and one
> solution I
> >> >> have found suggests concatenating the data from the 6
> sessions and
> >> >> including 2 regressors, one for the session number (i.e.,
> >> >> 1..2..3..4..5..6) and one for the transition period
> >> (specifying the
> >> >> last time point in a session and the first time point in the
> >> >> following session).
> >> >> Is it necessary and/or appropriate to do this?
> >> >
> >> >yes and no
> >> >
> >> >- do concatenate the sessions.
> >> >- create additional block-type regressors for the number
> of runs - 1
> >> >
> >> >you can make the regressors easily with the kron
> function. So if you
> >> >had runs = 6 scans = 100 (number of scans per run)
> >> >
> >> >r = kron(eye(runs-1),ones(scans,1));
> >> >
> >> >
> >> >>It seems more
> >> >> appropriate and straightforward to take the mean of the
> >> time series
> >> >>from each session for each of my VOIs to include in the
> >> analysis in a
> >> >>way similar to the PPI approach. However, this was not easy to
> >> >>implement within the DCM architecture within SPM5.
> >> >
> >> >with PPI the only reason to concatenate the runs is to setup
> >> the entire
> >> PPI
> >> >at one go. otherwise you can do what you did and run the
> ppi on each
> >> >run separately and put them all into a design later. you
> >> cannot do it
> >> >this way with dcm (i guess you could analyze each run
> >> separately, but
> >> >if each run
> >> has
> >> >a different trial mix or too few trials you might not get an
> >> >appropriate result).
> >> >
> >> >
> >> >darren
> >>
> >===========================================================
> ==
> >> ==========
> >> >==
> >>
> >>
> >===========================================================
> ==============
>
>
|