Dear Igor,
Well that's the question isn't it. i.e. there isn't a right way. IMHO I
would say that you have to try all the ways and see which one gives you
the biggest effect size ie. the best statistically significant
separation of the two groups regardless of your previous SPM analysis.
If you don't use the SPM clusters as your definition of the region then
you are back to the age old ROI vs voxel-wise debate. If you do use the
SPM clusters then as you point out you can't give them a categorical
anatomical label. You could rephrase the question by applying a ROI
template (e.g. of known affected areas) to the SPMT maps and then do a
logical discriminant analysis of whether a ROI contains significant
voxels or not, may be weighted for no. of voxels (or not) depending on
which gives you the best discrimination.
Good luck
HAPPY NEW YEAR.
Anna
Dear SPMers,
a categorical comparison between 2 groups of patients and controls (FDG-
PET; SPM2) produced statistically significant clusters within four
anatomical regions. I'd like to know metabolism in which anatomical
regions allows for the best
discrimination between controls and each of 2 patient groups. For this
purpose, i'm going to extract (individual) regional metabolic activities
from the regions and introduce them into a logistic regression/stepwise
linear discriminant analysis.
What is the most rational/valid way to extract regional activities? a)
ideally (?) would be to extract values exactly from the clusters
assigned to the definite anatomical regions (e.g. using a VOI tool of
SPM2). But the problem here is that 3 clusters assigned to the
anatomically different regions are merged with each other. I.e. the VOI
tool will
extract average values within 3 merged clusters (assigned to 3 different
regions) and 1 separate cluster. c) to extract individual values from
one peak (with maximal t-value) voxel or sphere of say 10mm within each
of 4 clusters again using the VOI tool.
b) to extract average values within the whole anatomical region (from 4
regions), where clusters were detected. Stereotactical binary masks are
needed, a marsbar tool can be used.
What would you advise?
Many thanks, Igor
*********************************** Igor Yakushev, MD Department of
Nuclear Medicine Johannes Gutenberg University Mainz Mainz/Germany
--
Anna Barnes, PhD
Research Associate (+441223 764676)
Brain Mapping Unit (Sir William Hardy Bldg.)
Dept. of Psychiatry
University of Cambridge
Downing Site
Downing Street
Cambridge, CB2 3EB
|