Hello Shary,

You mean a reference for the theory of second-level analyses? In a
neuroimaging context, two immediately come to mind. One is a Andrew
Holmes' HBM abstract (I believe it is from '98) and the other is a
Neuroimage paper with Karl as one (or perhaps the first) author
from a few years ago. The idea of the second paper was to compare
modeling random subject and fixed population effects as opposed to
using a summary image approach to population inference, the latter
approach being what most of us on the list call a second-level
analysis. (The reason for the vague citations is that I do not have
access to my reference library from my current location. I will send
on the complete references for these two papers to the list next
week).


Eric

Quoting "(Mr.) Shahryar (Shary) Rafi-Tari" <[log in to unmask]>:

> Hi Eric,
>
> I am also doing second level analysis.  But, I wonder what is the
> best reference for it so that I can refer to it in the paper I am
> writing.
> I appreciate it if you could give me a hint.
>
> Thanks,
> Shary
>
>
> At 13/01/2006 Friday 01:32 AM, Eric Zarahn wrote:
> >Cristina,
> >
> >(Your question concerned the correct "model", I presume you did
> not
> >mean preprocessing, segmentation, etc.) You have acquired two
> >images per subject, and so the appropriate estimator of the
> >population effect from each subject is simply the difference of
> the
> >standard space GM images derived respectively from the two
> acquired
> >images, say follow-up GM image minus baseline GM image. [As an
> >aside, the issue of whether you'd modulate with the Jacobian
> would
> >depend on whether you want the voxel-wise measures to be of
> local
> >GM concentration (no modulation) or local GM volume (modulation)
> in
> >standard space.] You could take this difference using imcalc or
> some
> >other suitable MATLAB routine. These difference images would
> then be
> >entered as the dependent variable in a population level model as
> you
> >describe in your step 2.
> >
> >Eric
> >
> >Quoting CrisTesta <[log in to unmask]>:
> >
> > > Dear SPMers,
> > > I have structural MR images of two AD groups (treated and
> > > placebo) acquired at baseline, and follow up.
> > > I processed my longitudinal data following Chetelat's
> protocol
> > > (Neuroimage 2005) with SPM2.
> > > Now I am interested in a second level analysis. I'd like to
> > > compare GM tissue loss in the placebo group vs GM tissue loss
> in
> > > the treated group, including age and gender as nuisances.
> > > If I correctly understood, I have to:
> > > 1) determine one con*.img for each subject;
> > > 2) and then compare con*.img of subjects A with those subj B,
> > > with a simple t-test ("compare population" model)
> > >
> > > Please, could you tell me the correct model I should use for
> the
> > > first step (i.e. to determine one con*.img for each subject,
> > > taking into account that these are longitudinal data)?
> > > Thank you in advance
> > >
> > > Cristina
> > >
> > > LENITEM - Laboratory of Epidemiology Neuroimaging &
> Telemedicine
> > > IRCCS San Giovanni di Dio FBF - The National Center for
> Research
> > > and Care of Alzheimer's Disease
> > > via Pilastroni 4, 25125 - Brescia, Italy
> > > Tel: +39 030 3501 361
> > > www.centroAlzheimer.it
>