Joe,
I agree that the session question is best answered empirically. My sense
though is that session is a highly artificial variable, and one that may
have more or less meaning depending on the scanner/research set up. For
the data that I am currently analyzing the amount of time that passed
between sessions was on average about 15s, enough time to ask the subject
if he was ok and set up for the next experimental block. The sessions
themselves lasted 6 minutes, but they could have lasted 8 or 2 -- the
decision was more or less arbitrary. That said, it seems to me that both
the state of the subject and the state of the scanner change much more
within session (6 minutes being much larger than 15s) than between session
and that moreover these "states" fluctuate continuously. There may be
nothing about "session 1" that substantively distinguishes it
from "session 2" in the way that subject 1 can be uniquely distinguished
from subject 2 -- unless it can be shown there exists a tell-tale session
parameter that varies not continuously but abruptly between sessions, as,
for instance, IQ changes between subjects. (Of course, motion correcting
only within session would tend to create an artifical difference between
sessions.)
My other question was about how flame's strategy regarding parameter
estimates (different for near-threshold voxels) affects analyses at higher
levels where the set of near-threshold voxels is different?
thanks,
Brad Buchsbaum
>If this is correct, then I am
>a little unsure in what sense "session" (other than for signal
processing
>reasons) is given such importance. It is obvious why "subject" is an
>important variable, one that explains a lot the variance. Different
>subjects have different intrinsic properties, etc. But can the same be
>said about a "session"? For one, sessions are not really things, they
are
>temporal demarcations, so a given session can't really have intrinsic
>properties.
Actually, in fmri sessions do have intrinsic properties due to the
physical
state of the scanner (which changes over time) and the physical state of
the participant. In addition, any head movement within the field (say
in
between two separate "runs") affects the data by altering the
effectiveness
of the shim and moving the head with respect to the region of maximal
magnetic field homogeneity -- neither of which is corrected by "motion
correction." I guess the big question is "how important are these
factors
compared to the differences across subjects?" and "what evidence is
there
for these issues?" There was a recent paper in MRM by Aguirre's group
looking at BOLD and perfusion (ASL) variability across sessions within
and
between days which is relevant but that's the only one I can think of at
the moment. Perhaps someone could suggest others?
Joe
|