I think this is important to note.
So do you see any problems in applying FLOBS or filmbabe at the first level with regards to higher-level stats?
Von: Mark Woolrich [mailto:[log in to unmask]]
Gesendet: Mi 24.11.2004 12:49
An: [log in to unmask]
Betreff: Re: [FSL] FLOBS?
>I am looking for a method to test for differences in the BOLD-time course
>between two groups. Ideally, I would like to test for differences in time
>to peak, full width half maximum, time to onset etc.
>Perhaps FLOBS can be used to do this.
Unfortunately, not easily. You probably have in mind the parameterised HRF
made up of half cosines upon which FLOBs is based. However, FLOBS only
uses this parameterised HRF to generate a linear basis set. Hence any
explicit parameterisation of time to peak etc. is lost. It is possible to
recover the time to peak etc. by extracting them from the inferred basis
set impulse response function (IRF) (e.g. finding the time to peak for the
smoothed basis set IRF is not difficult).
>I have one condition only: when I
>use FLOBS using the default functions, I can ask for 3 contrasts (one for
>each regressor) and for the F-test giving me a linear combination of the
>My question now is: how can I use these FLOBS results at second level?
>Are there ways to compare the best fit of the BOLD response between group
>A and B?
>Is there perhaps a way to extract measures such as time to peak, fwhm
>etc, and then use these at second level?
This relates to a general problem that exists when we try to use basis
sets with group studies. That is, if we pass up the COPEs for each of the
basis functions to a second level, and then look for differences between
groups of subjects, then we could get a significant difference not only
for a difference in size of effect, but also for a difference in
shape of effect. There is no simple way to separate out this two
possiblities using basis functions. Instead, you would need to recover
explicit estimates of size, time to peak etc. from the inferred basis set
IRF from the output from the first level (see above) and then
pass them to the second level. There is then the question of whether you
can assume gaussianity at the second level for these sorts of parameters,
I have no experience of this - so couldn't say definitively. If they are
Gaussian then you can do a traditional OLS random effects analysis, if not
then you could still use nonparametric stats.
Oxford University Centre for Functional MRI of the Brain (FMRIB),
John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK.
Tel: (+44)1865-222782 Homepage: http://www.fmrib.ox.ac.uk/~woolrich