Dear SPMers,
I have posted a variant of this question last month,
but require some advice on charecterising learning
related changes across several sessions.
Conventional box-car designs would assume that each
block within experiment is same in terms of BOLD
response, which is contrary to the neuroimaging
literature. Several neuroimaging studies use
parametric modulation with time as a way of teasing
out learning-related changes, but i guess this model
essentially assumes linear increase (or decrease) in
brain activation with learning. This is NOT true in
case of behaviour which is usually quite steeper in
the beginning of learning and saturates after learning
with a profile like a bounded exponential function. I
guess that instead of using such models, using
Response Times as parametric modulators is more
appropriate as it tracks the behaviour.
We have a block design experiment with four sessions
for each subject. Learning progresses over the entire
duration of experiment. In a session separable model,
we enter response times for every block (replicating
the same number for all scans within the block) as
user specified regressor after convolving with
canonical hrf (spm_hrf). Is this equivalent to a
parametric modulation for assessing learning related
changes.
I would guess that a user specified regressor gives
more flexibility of specifying variations across
sessions, while parametric modulation does not. Is
this correct?
A further question is whether Response Times can be
used as covariate in first level design matrices as it
is an observed effect and is hence not an explanatory
variable. The argument is that by using performance
measure such as Response Times as a covariate, we
would be testing for learning related activations.
Using a user specified regressor/parametric modulator
and a box-car in the same design matrix also implies
that these are not orthogonal to the box-car. Is this
correct? If yes, would this prevent from using linear
contrasts to test for box-car related or performance
related BOLD responses.
A final question is if using a linear contrasts in a
fixed effects fashion (i.e. specifying a 1 in contrast
vector) for box-car and regressor/parametric modulator
be equivalent to box-car only design matrix.
Any insights are greatly appreciated
Thanks in advance
-krishna
Krishna P. Miyapuram
Work : Graduate Student, Dept of Anatomy, Univ of Cambridge
Home : 25, Parkside, Room 10 (Phone +44 - 1223 - 476858
Address : Downing College, Regent Street, Cambridge, CB21DQ, England
URL : http://people.pwf.cam.ac.uk/kpm23/index.html
__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
|