Dear Allison,
>Dear SPM experts:
>
>I am trying to use SPM 99 to do a group analysis of some fMRI data
>concerning cigarette smoking. Subjects are initially at rest in the
>scanner (e.g. for 15 images), and then begin inhaling smoke (eg. for 85
>images). We would like to know which areas of the brain increase their
>activity as nicotine blood levels rise. Since we don't yet have a
>catheter in place to directly measure blood levels, we will approximate it
>as a linear increase from the time of first inhalation.
>
>How should I ask SPM to do a linear regression of this kind?
>
>Some more specific questions that have arisen from reading other
>messages:
>
>Is this considered a session x condition interaction, such that
>scans from all subjects need to be entered as one session?
>
No, different subjects should definitely be different sessions.
>If I omitted the resting images, focusing only on the smoking period,
>would I be able to use the 'time' option in the parametric modulation
>choices?
>
Since you will have, sa far as I understand, only one "epoch" per session
you will not be able to use parametric modulation anyway.
>Should I treat the smoking period as an epoch, and if so what type of
>response should I assign it?
>
Your design is certainly quite unorthodox, and I am sure there are a lot of
suggestions out there on how to model your data. One suggestion that comes
to mind is as follows.
Model smoke inhalation as an epoch, fixed SOA and epoch lengths, the length
of being 85 TR's (if there are 85 images sampled during inhalation, I
assume here inhalation continue until cessation of session). This will
model any effects with an immediate onset (e.g. the smell of smoke, feeling
of well being etc.). In addition put in a user specified covariate being
zeros for the scans with no inhalation, followed by a linear increase for
the remaining scans (e.g. [zeros(1,15) 0:85]). Obviously you should not use
a high pass filter.
An F-contrast for effects of interest will now give you any regions which
respond to smoke inhalation/nicotion concentration, whereas a t-contrast
with a one for your user specified covariate will give you areas which
incerase lineraly with nicotine concentration.
Now, this suggestion suffers from a number of weaknesses, e.g. the
step-response and the linear ramp are going to be highly correlated, which
means your model will be extra sensitive to deviations from your assumption
of linearly increasing nicotine levels in the blood.
Also, inherent in your design is a confounding of time and nicotine
concentration which may render the interpretation slightly tricky.
However, I hope the suggestion above will give you a starting point. Maybe
you should run the analysis on one subject first, identify regions with
high F-values, plot data from those regions and use those time courses to
formulate a "better" hypothesis for a group analysis of the remaining
subjects.
>Thank you in advance for your help!
>
>Warm regards,
>Tavis
>
>
Good luck Jesper
Jesper Andersson
Wellcome Dept. of Cognitive Neurology
12 Queen Square
London WC1N 3BG
phone: 44 171 833 7484
fax: 44 171 813 1420
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|