Just an addendum: in SPM5 (unlike SPM2), the regressors for each basis
function are also serially orthogonalised AFTER convolution with your
events/epochs (corresponding to Eric's approach (a) below), so the problem
that Jesper identified when testing only the canonical HRF in an epoch model
that also employs derivatives, no longer arises.
However, I would also add that Eric's two suggestions are only appropriate
if the canonical HRF is a good model of the BOLD response (as he says). If
this is not the case, the first-order Taylor expansion that underlies both
approaches (a)+(b) breaks down. Thus in the more general case (eg with a set
of gamma functions, or Fourier or FIR in SPM), one does not regard basis
functions as "nuisance" variables - they all carry equal weight, and are
best tested via F-tests.
Rik
>-----Original Message-----
>From: SPM (Statistical Parametric Mapping)
>[mailto:[log in to unmask]] On Behalf Of Eric Zarahn
>Sent: 16 March 2006 23:37
>To: [log in to unmask]
>Subject: Re: [SPM] time derivative for block design fMRI???
>
>
>Hi Matthew and Yanmei,
>
>
>
>Just to add, the last comment in that cited post:
>
>
>
>"Having said all this. It is still *BETTER* to use an F-test
>whenever you
>are encoding a condition using more than one basis function (the reason
>for explained above). And since we can now (as of proper variance
>component estimation) bring all the parameter estimates (or rather
>contrasts thereof) to the second level, it is *always* the recommended
>thing to do."
>
>
>
>is not correct. You can dilute your statistical effect size
>(regression sums
>of squares) across the multiple dimensions of an F-contrast.
>It is better to
>use a one-dimensional contrast when you know what you are looking for
>(matched filter idea). So, for time-series statistical
>inference I would
>suggest using either (a) a one-dimensional contrast estimating
> the original
>effect in the presence of orthogonalized nuisance terms such
>as derivatives,
>as Jesper mentioned in the model (Zarahn, 2002), or even better (b) an
>estimate of amplitude based on both the original and derivative terms
>(Calhoun et al, 2004). As an aside, approaches like (a) don't
>matter to 2nd
>level analyses, while (b) will affect 2nd level estimators
>(beneficially)
>
>
>
>Eric.
>
>
>
>
>
>Calhoun, V.D., Stevens, M.C., Pearlson, G.D., Kiehl, K.A., 2004. fMRI
>analysis with the general linear model: removal of
>latency-induced amplitude
>bias by incorporation of hemodynamic derivative terms.
>Neuroimage 22, 252-7.
>
>
>
>Zarahn, E., 2002. Using larger dimensional signal-subspaces to
>increase
>sensitivity in fMRI time series analyses. Human Brain Mapping
>17, 13-16.
>
>
>
>
>
>----- Original Message -----
>From: "Matthew Brett" <[log in to unmask]>
>To: <[log in to unmask]>
>Sent: Thursday, March 16, 2006 5:11 PM
>Subject: Re: [SPM] time derivative for block design fMRI???
>
>
>Hi,
>
>> I have a very naive question about inclusion of time derivative in
>> analysing block design fMRI data (TR = 2s, block duration =
>20s). Is it
>> suggested to include time derivative for block design or it
>is only used
>> for event-related?
>
>I think the answer to that would be that it is basically only used for
>event-related. There is a characteristically clear explanation from
>Jesper Andersson here:
>
>http://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind04&L=SPM&P=R35
1987&I=-3
Best,
Matthew
|