Dear group
Does anyone know if fMRI data processing with spm99 is in any sense faster
than with spm96?
Thanks
Jeff Lorberbaum
Jeff Lorberbaum, MD
Medical University of South Carolina
On Wed, 14 Feb 2001, Karl Friston wrote:
> Dear Kelvin,
>
> > I'm reading your paper on how to correct for threshold according to the
> > spatial extent of the activation. However, I got some question that I
> > hope you could give me some advice.
> >
> > We're dealing with a group map problem and we'd like to average the
> > statistical map of say 12 patients (corrected by dividing by square
> > root of 12).
> >
> > The problem arise when we're trying to calculate a mean contrast map to
> > show if there's any activation in doing a specific task. Generally,
> > we'd like to correct for the blob size, that'd produce different
> > threshold for different dataset. Although we could threshold each
> > dataset with individual threshold and averaged the thresholded SPM,
> > we'd like to find out if it'd be better to correct the blob size after
> > we averaged the pre-threshold SPM.
> >
> > It'd be great if you could explain a bit on why you choose which approach.
>
> I would suggest that you do not average the subject-specific SPMs
> created by subject-specific contrast weights but create one SPM based
> on the subject-specific contrast images. This corresponds to a
> second-level or random effects analysis of the group activation. It is
> simply implemented by modeling all the subjects in a first-level model
> in a subject-separable fashion. Create 12 subject-specific contrast
> images con???.img (each testing for the activation in each subject).
> Finally enter these 12 contrast images into a one sample t test in
> 'Basic Designs'. The inference based on spatial extent can now be
> applied to the ensuing SPM{t}.
>
> In general you do not average statistics but derive a single statistic
> based on the average.
>
>
> I hope this helps - Karl
>
|