Alex,
I have been working on LORETA/PET comparisons in the auditory system.
One problem I came across was that the time scale the two systems
typically epoch over is very different. I am getting around this by
using steady state responses. I'm not sure what the time scale of the
potential you are measuring is, but It may be something to consider if
you EEG epoch is considerably shorter than your PET activation.
I also have some comments on your other questions:
- do you think the LORETA data could and should be analyzed using SPM?
Yes, but not in the same manner that you typically use for PET.
Some specific questions are:
- does SPM make assumptions that do not hold for LORETA?
One step I would not apply to LORETA images that I would to PET images
is a correction for global CBF.
- does the smaller whole-brain-volume of LORETA (2394 voxels) matter?
It means you have a much lower resolution image, but you knew that
already. It will decrease the number of comparisons that will be made
for that image. This will decrease the number size of the Bonnferoni
correction that is made in calculating the corrected voxel p value.
Smoothing the PET image considerably may allow you to have the same
number of resels, but then you are loosing much of the advantage PET
offers in terms of spatial resolution.
- any reasons why LORETA volumes would not conform to random field
theory?
LOERTA optimizes maximum smoothness of current density against the fit
of the data. This means that a LORETA image of pure noise will not look
like a PET image of pure noise. There is likely to be a large
correlation between adjacent pixels even when reconstructing pure
noise.
You will also want to consider the regularization factor you have
applied to balance the modeling vs.data parameters. If possible this
should be the same in all conditions (assuming SNR is constant).
Generally, as the regularization factor increases, the model term will
begin to dominate. This will make the image conform less and less to
random filed assumptions.
>From what I can gather, this all means that you cannot use cluster level
statistics when comparing LORETA images. The voxel level comparisons
are still valid.
- Is the Analyze-Format available so that LORETA images could be
transformed into an SPM-readable format?
I'm pretty sure this is in the docs. If you are using the Curry
software packages, I have some scripts that may be of use to in doing
this. Write me if you are interested.
One final thing to chew on is that LORETA finds a vector at each
position with three components. This provides three non-independent
views of the same event. One group has used the magnitude of this
vector to give them a single image. Another approach may be to perform
a MANCOVA with three measurements per event. I have been working on
figuring out how to do this in SPM, if you or anyone else on the list
can point me in the right direction I would be very grateful.
Best wishes,
---sam Reyes
PhD Student
SUNY Buffalo
Hearing Research Lab
e-mail: [log in to unmask]
Phone(s): (716) 829-2001; (716) 862-8790
Fax: (716) 829-2980
Mailing Address: 215 Parker Hall
Buffalo, NY 14214
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|