Dear Kris,
>
> I also need to explain where in the analyses of the fMRI images the
> intrinsic correction for temporal autocorrelation occurs. I've not been
> able to figure this out on my own. (sorry if this question is a little
> daft)
In SPM, the temporal autocorrelation is taken into account at the
parameter estimation stage and in the statistical inference.
The design matrix and the data are both convolved with a filter kernel,
which is usually a bandpass filter, i.e. it is effectively a combination
of a user-specified lowpass and highpass filter. This changes the
autocorrelation structure of the data such that the actual
autocorrelation structure is given by convolution of the (unknown)
intrinsic autocorrelation with the bandpass filter kernel. One goal of
this filtering is to impose an autocorrelation structure on the data,
which is not too different from the assumed autocorrelation
(s. below).
At the level of statistical inference: to compute a t-value at each
voxel, one has to estimate the intrinsic autocorrelation of the data.
Currently, in SPM99, you can do that by assuming that the intrinsic
autocorrelation before the convolution with the bandpass filter kernel
is a unity matrix or by estimating the autocorrelation with an
AR(1)-model. Anything what follows at this stage, e.g. the computation
of the effective degrees of freedom is based on these estimates of the
intrinsic and actual autocorrelation structures.
Some part of all this is described in
KJ Worsley and KJ Friston, 1995. Analysis of fMRI Time-Series Revisited
- Again.
Neuroimage, 2:173-181
As far as I know, a paper by Karl Friston et al. about temporal
filtering is in press (NeuroImage).
Stefan
--
Stefan Kiebel
Functional Imaging Laboratory
Wellcome Dept. of Cognitive Neurology
12 Queen Square
WC1N 3BG London, UK
Tel.: +44-(0)20-7833-7478
FAX : -7813-1420
email: [log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|