Dear Karl & Tom,
> PS: I have a vague memory that someone has already looked at this.
> Perhaps they may respond to this enquiry?
I think that was us two, Karl ... ;-)
We briefly discussed the issue of the field-dependency of the
biophysical constants and its potential effect on parameter
estimation about a year ago, and I have been planning to look at this
issue ever since then. So thanks for the reminder, Tom!
In fact, there is some work by Liu et al. that offers adjusted
constants for the Balloon model at 3T:
Liu, T. T., Luh, W. M., Wong, E. C., Bandettini, P. A., Obata, T.,
Frank, L. R., and Buxton, R. B. 2000.
On the nonlinear relation between BOLD and CBF.
Proc. Intl. Soc. Mag. Reson. Med. 8: 948.
Using these values and the simulation machinery we have set up for
DCM, we could perform some rather straightforward evaluations of the
robustness of DCM wrt. the exact values of these constants. For
example, we could generate data, using the 3T constants as described
by Liu et al., add observation noise and compare whether the ability
of DCM to reconstruct the true coupling parameters differs
significantly depending on whether one uses 1.5T or 3T
constants. Also, we could take an empirical 3T data set and (i) test
whether the use of different biophysical constants leads to
significant differences in the estimates of the coupling parameters
and (ii) to what degree a model based on the 3T constants leads to a
larger model evidence than the same model based on 1.5T constants.
If you think this is a useful strategy, Karl, I will implement the
above procedures asap. Would you like to join us in this endeavour,
Tom? I have also discussed these ideas with Nikolaus Weisskopf who
would be interested to contribute to this project if we go ahead.
Very best wishes,
Klaas
At 11:51 18/07/2005, Karl Friston wrote:
>Dear Tom,
>
>I have to confess we have been using the same constants
>in the output nonlinearity for both our 1.5 and 3.0T data.
>
>The reason is that there are sufficient conditional dependencies
>among the [uninteresting] biophysical parameters for each region
>to fit the data and these parameters are largely (conditionally)
>independent of the [interesting] coupling parameters.
>
>Put simply, as long as the hemodynamic part of the DCM is flexible
>enough to model each regional HRF accurately, the conditional
>density of the coupling parameters should be fairly stable.
>The current priors provide a quite flexible model. Note that the
>hemodynamic part does not have to be biophysically realistic (although
>the DCM model is reasonably realistic).
>
>Having said this, we have never checked the impact of changing
>the output constants and field strength. This would be an interesting
>exercise that may motivate making these constants free parameters
>with relatively tight priors. I would only do this if you were interested
>in this issue (and wanted to write a short technical note on it).
>
>I hope this helps - Karl
>
>PS: I have a vague memory that someone has already looked at this.
>Perhaps they may respond to this enquiry?
>
>
>
>At 13:59 15/07/2005 +0200, you wrote:
> >Dear Dr. Friston,
> >
> >I would like to conduct my next DCM study on a 3T system. What worries
> >me is that the coefficients in the Balloon model are valid for 1.5 T and
> >a TE of 40 ms. In your methodological paper on DCM you show that
> >violations of priors on hemodynamic or biophysical parameters (in a
> >certain range)does not greatly affect normalized coupling parameters. Do
> >you think that DCM is robust changes in T2* and Intra/Extravascular
> >contributions in gradient echo BOLD induced by moving from 1.5T to 3T or
> >would it be necessary to evaluate these changes first and modify the
> >coefficients accordingly?
> >
> >Sincerely yours
> >
> >Tom Ethofer
> >
> >
> >
|