Dear Pierre,
Dear SPM'ers,
I have some questions concerning the basic "philosophy" of path analysis.
The point is applying path analysis to covariance matrix obtained from two
subject groups or two different conditions within a group.
When path analysis is applied to two groups of subjects (for an example a
control group and a schizophrenic group) and shows that the models for the
two groups are different, I think we can conclude that the "structural"
connections are different for the two groups, i.e. the "structure of the
brain" is different (that does not for me necessarily imply that an
anatomical difference exists).
I think you are wrong in concluding that the models are demonstrating a
group difference in structural connections (I don't exactly know what you
mean by this, but I assume it refers to either an explicit anatomical
difference, or something akin to differences in the number of axons
projecting from area A, let us say, to area B, or something about the
synaptic weights, etc). It is true that a group difference in structure
could lead to a group difference in the functional network; but a group
difference in functional networks could also mean a functional difference in
how the subjects do the task (e.g., fewer dopamine receptors may lead to
functional difference; now, it is purely definitional whether you would want
to count this as a structural difference or not; different strategies;
etc.).
When path analysis is applied to different tasks performed by a subject (or
an homogeneous group of subjects), a "good" model have to fit the
covariance matrix obtained during all the tasks. If not (imagine that
during a visual task the path coefficients are different from the
coefficients in a rest condition), I cannot suppose that the "structure of
the brain" is modified within the few seconds between the two tasks. I
suppose that if the model is valid during active condition and not during
rest, this means that either some areas have been omitted in the model or
that the variance for each areas is very low in the rest condition
(variations lower within rest than active condition) leading to low
efficiency of the path analysis.
If you are examing two tasks (I think rather than a relatively uncontrolled
condition such as rest, think of two tasks such as identifying objects
versus identifying where the objects are in space), then what path analysis
demonstrates is that tthe same anatomical network (the regions in visual
cortex included in your anatomical model) is supporting two functional
networks -- one for object processing, one for spatial processing. Each
anatomical link may have different effective connections depending on the
task.
I think your interpretation is correct if one task fits the model, and one
doesn't.
I think that we have to calculate the covariance matrix using all the
condition and search for models that explain this covariance matrix, i.e. a
"structural" model has to be true whatever the task. Do somebody subscribes
or subscribes not to this?
In normal subjects, we, in principle, know the structural model that is true
whatever that task -- it is the central nervous system. What you are
probably saying is that one should use a set of tasks that are likely to
employ a given anatomical network, and then demonstrate that each task has a
different set of interregional effective connections, and thus each task
uses a different functional network within the same anatomical matrix. This
was the original way in which path analysis was employed in functional brain
imaging. But, I would think that it would be useful if one demonstrates
that one task does have a good fit to the network, whereas another, which
one might think should be implemented by the anatomical network, doesn't fit
the network.
Cheers,
Barry
________________________________________________________
Barry Horwitz, Ph.D.
Senior Investigator
Language Section
Voice, Speech and Language Branch
National Institute on Deafness and other Communication Disorders
National Institutes of Health
Bldg. 10, Rm. 6C420
MSC 1591
Bethesda, MD 20892
USA
Tel. 301-594-7755
FAX 301-480-5625
[log in to unmask]
http://www.nidcd.nih.gov/intram/scientists/horwitzb.htm
|