Hello Todd,
Just to add something that perhaps you are already aware of: that
the inclusion of the derivatives themselves provides some latitude
for the parameters the derivative was taken with respect to (that's
indeed their purpose for inclusion) Thus, inclusion of temporal and
dispersion derivatives means the model will do a reasonable job of
modeling effects due to an hrf with a lag and dispersion in a
neighborhood of the canonical hrf (or whichever hrf you specify).
The approximation will not be good for large deviations from the
canonical hrf. From the coefficients on the derivatives you can
estimate the new lags and dispersions.
Sincerely,
Eric
Quoting Will Penny <[log in to unmask]>:
> [log in to unmask] wrote:
>
> >
> >
> > Sorry, one other quick question. The canonical HRFs and
> gamma
> > function regressors are built in (it appears) so their
> > delays/dispersions cannot be adjusted manually?
>
> You can change these by passing different parameters to spm_hrf.m
>
> These options are not available from the GUI.
>
> I know that you can
> > include derivatives with the canonical HRF, but it only gives
> you two
> > derivative functions with built-in delays/dispersions that
> can't be
> > adjusted from what I can see. If you want to customize your
> own gamma
> > functions or canonical HRFs, you need to build them yourself
> right?
> >
>
> Yes. But I would'nt bother for a block design.
>
> Best,
>
> Will.
>
> > Thank you,
> >
> > Todd Penney
> >
> >
> > Quoting Will Penny <[log in to unmask]>:
> >
> >> Dear Todd,
> >>
> >> Nothing really strikes me as unusual.
> >>
> >> All I can think is - maybe its something to do with the
> >> Gamma basis functions, so try the canonical HRF.
> >>
> >> This will probably make no difference.
> >>
> >> Did you save a copy of the model specification job file ?
> >> If so, perhaps you could mail it to me (not to the list).
> >>
> >> Otherwise - you could go into Review->Explore Design,
> >> plot the design matrix and send the image to me.
> >>
> >> This may provide some more clues.
> >>
> >> Best,
> >>
> >> Will.
> >>
> >> [log in to unmask] wrote:
> >>
> >>>
> >>>
> >>>
> >>> Hi Will,
> >>>
> >>> You are correct - I am doing a 1st level fMRI design as I
> only have
> >>> data from one patient from one session. I am using a 1st
> order Gamma
> >>> function as my model with a window length of 20sec. I fixed
> the
> >>> original problem I was having entering contrasts in the
> contrast
> >>> manager. The fix was available in the SPM archives, which a
> fellow
> >>> from Harvard pointed me to. I am using Windows XP to run
> SPM5 and
> >>> for some reason the command
> >>>
> >>> feature('JavaFigures',0)
> >>>
> >>> needs to be run at the MATLAB command prompt to correct the
> problem
> >>> with the contrast manager I was having. Now it will allow me
> to
> >>> enter contrasts properly.
> >>>
> >>> The contrast vector I am using is [1 0] because I only have
> two
> >>> regressors and I only want to see the effect of the first
> one.
> >>>
> >>> ANOVA, as far as I can tell from the SPM5 manual, is used to
> solve
> >>> for the covariates when you ask for a Bayesian analysis of
> your
> >>> model, not when you want the classical method of solving the
> GLM
> >>> applied. Anything that strikes you as incorrect with what
> I'm doing?
> >>>
> >>> Thanks
> >>>
> >>> Todd Penney
> >>>
> >>> Quoting Will Penny <[log in to unmask]>:
> >>>
> >>>>
> >>>>
> >>>> [log in to unmask] wrote:
> >>>>
> >>>>>
> >>>>>
> >>>>> Hi Will,
> >>>>>
> >>>>> I don't see anywhere in SPM5 where you specify ANOVA with
> or
> >>>>> without a constant model.
> >>>>
> >>>>
> >>>>
> >>>> SPM5 makes ANOVA models without a constant.
> >>>>
> >>>> The model functions I am using are Gamma functions if
> >>>>
> >>>>> that interests you.
> >>>>
> >>>>
> >>>>
> >>>> OK. So this is a first level fMRI design.
> >>>>
> >>>> What is the model and what contrasts are you trying to enter
> ?
> >>>>
> >>>> [But you can actually type them in ? - as there was recently
> a bug
> >>>> where you could'nt even do this ...]
> >>>>
> >>>> Best,
> >>>>
> >>>> Will.
> >>>>
> >>>> I don't see anything about ANOVA anywhere though.
> >>>>
> >>>>> Any thoughts?
> >>>>>
> >>>>> Thanks for your help.
> >>>>>
> >>>>> Todd Penney
> >>>>>
> >>>>>
> >>>>> Quoting Will Penny <[log in to unmask]>:
> >>>>>
> >>>>>> Dear Todd,
> >>>>>>
> >>>>>> What is the contrast you are trying to enter ?
> >>>>>> And what is the model ?
> >>>>>>
> >>>>>> Some contrasts are invalid.
> >>>>>>
> >>>>>> For example if you have three conditions and use the ANOVA
> with
> >>>>>> a constant model, SPM will make a design matrix with
> >>>>>> 4 columns. The first three for condition effects and the
> last for
> >>>>>> the overall mean.
> >>>>>>
> >>>>>> The trouble with this model is that it is
> over-parameterised.
> >>>>>> Each column in the design matrix can be gotten from a
> linear
> >>>>>> combination of the other three. So there is
> >>>>>> some indeterminacy in the associated regression
> coefficients.
> >>>>>>
> >>>>>> The end result is that contrasts like 1 0 0 0
> >>>>>> or 0 1 0 0 are not admissable because the associated
> regression
> >>>>>> coefficients are not interpretable.
> >>>>>>
> >>>>>> But you can look at differences in this model eg 1 -1 0 0
> >>>>>>
> >>>>>> If you wanted to look at each effect alone you'd use
> >>>>>> a different model eg. ANOVA without a constant. Then
> >>>>>> there's no problem.
> >>>>>>
> >>>>>> (Basically, the ANOVA with a constant is a silly model).
> >>>>>>
> >>>>>> Anyway, what model do you have ?
> >>>>>>
> >>>>>> Best,
> >>>>>>
> >>>>>> Will.
> >>>>>>
> >>>>>>
> >>>>>>
> >>>>>>
> >>>>>>
> >>>>>> Todd Penney wrote:
> >>>>>>
> >>>>>>> Hello again all,
> >>>>>>>
> >>>>>>> I've tried entering the contrasts several different
> ways, paying
> >>>>>>> careful
> >>>>>>> attention not to exceed the number of valid contrasts
> based on
> >>>>>>> the number of
> >>>>>>> conditions I have. This problem is still persisting.
> I'm
> >>>>>>> beginning to think
> >>>>>>> there is some problem with the way SPM5 is set up on my
> PC. Any
> >>>>>>> suggestions
> >>>>>>> from SPM experts?
> >>>>>>>
> >>>>>>> Thank you
> >>>>>>>
> >>>>>>> Todd Penney
> >>>>>>> University of Alberta
> >>>>>>>
> >>>>>>> ----- Forwarded message from [log in to unmask] -----
> >>>>>>> Date: Wed, 15 Feb 2006 11:55:36 -0700
> >>>>>>> From: [log in to unmask]
> >>>>>>> Reply-To: [log in to unmask]
> >>>>>>> Subject: problems setting up contrasts using SPM5
> >>>>>>> To: [log in to unmask]
> >>>>>>>
> >>>>>>>
> >>>>>>> Hello,
> >>>>>>>
> >>>>>>> I'm at the step of setting up contrasts to review my
> results
> >>>>>>> using SPM5. In
> >>>>>>> the contrast manager when I attempt to put in a contrast
> vector I
> >>>>>>> get an error
> >>>>>>> saying "No valid contrast vector defined". I've tried
> every way
> >>>>>>> of putting in
> >>>>>>> the vector using quotes, brackets, commas between
> elements, etc.
> >>>>>>> However
> >>>>>>> nothing seems to work. The contrast is named also, so
> there is
> >>>>>>> no problem with
> >>>>>>> that. Has anyone had a similar problem?
> >>>>>>>
> >>>>>>> Thank you
> >>>>>>>
> >>>>>>> Todd Penney
> >>>>>>> University of Alberta
> >>>>>>> Edmonton, Alberta, Canada
> >>>>>>>
> >>>>>>> ----- End forwarded message -----
> >>>>>>>
> >>>>>>>
> >>>>>>> ----- End forwarded message -----
> >>>>>>>
> >>>>>>>
> >>>>>>
> >>>>>> --
> >>>>>> William D. Penny
> >>>>>> Wellcome Department of Imaging Neuroscience
> >>>>>> University College London
> >>>>>> 12 Queen Square
> >>>>>> London WC1N 3BG
> >>>>>>
> >>>>>> Tel: 020 7833 7475
> >>>>>> FAX: 020 7813 1420
> >>>>>> Email: [log in to unmask]
> >>>>>> URL: http://www.fil.ion.ucl.ac.uk/~wpenny/
> >>>>>>
> >>>>>>
> >>>>>
> >>>>>
> >>>>>
> >>>>>
> >>>>
> >>>> --
> >>>> William D. Penny
> >>>> Wellcome Department of Imaging Neuroscience
> >>>> University College London
> >>>> 12 Queen Square
> >>>> London WC1N 3BG
> >>>>
> >>>> Tel: 020 7833 7475
> >>>> FAX: 020 7813 1420
> >>>> Email: [log in to unmask]
> >>>> URL: http://www.fil.ion.ucl.ac.uk/~wpenny/
> >>>>
> >>>>
> >>>
> >>>
> >>>
> >>>
> >>
> >> --
> >> William D. Penny
> >> Wellcome Department of Imaging Neuroscience
> >> University College London
> >> 12 Queen Square
> >> London WC1N 3BG
> >>
> >> Tel: 020 7833 7475
> >> FAX: 020 7813 1420
> >> Email: [log in to unmask]
> >> URL: http://www.fil.ion.ucl.ac.uk/~wpenny/
> >>
> >>
> >
> >
> >
> >
>
> --
> William D. Penny
> Wellcome Department of Imaging Neuroscience
> University College London
> 12 Queen Square
> London WC1N 3BG
>
> Tel: 020 7833 7475
> FAX: 020 7813 1420
> Email: [log in to unmask]
> URL: http://www.fil.ion.ucl.ac.uk/~wpenny/
>
>
|