Dear Tim,
>Sorry to bother you (and apologies if I missed a reply to Jane Warren about
>this on the helpline). I had understood that DCM could not be carried out
>on sparse data sets, like our typical sequence based on TR of 12s for
>auditory experiments. I am keen to try this out in some data sets we
>already have where we have very specific a priori anatomical hypotheses
>about connections between PAC, Planum Temporale and superior temporal sulcus.
DCM for sparsely sampled time-series is, in principle perfectly OK.
The generative model for DCM can predict data that are acquired sparsely
and therefore Bayesian inversion of the model can proceed with sparse data.
The key issue is the acquisition time - TA (not the repetition time - TR).
DCM assumes that each image is acquired instantaneously. Therefore, we
recommend that the TA is about 2 seconds. However this does not place
any bound on the TR
Clearly, sparse sampling will preclude measurements of high frequency
responses. This may reduce the conditional precision of coupling estimates.
However, this potential reduction is am empirical issue that could
be addressed with simulations (using the DCM simulation tool) or by
simply decimating a standard time-series. Of course, the simplest way
to address this issue is to analyse a sparse time-series and see if you get
good results :)
I hope this helps - Karl
|