On Tue, Nov 15, 2016 at 4:03 AM, Penny, William <[log in to unmask]> wrote:
>
> I don't know the answer to your question but I agree with your reasoning.
Thanks. I might try using the t-values as quasi-regularised parameters
for an encoding model, and see how it goes.
> Do you know about the Multivariate Bayes (MVB ) facility in SPM -?
> this may also be useful for you. Its designed to look at local representations
> in the brain (ie. not whole brain) but can ask questions like - is region X
> better than Y for decoding task A ? Is the representation in X sparse
> or distributed ?
Thanks. I know of that paper
http://www.fil.ion.ucl.ac.uk/~karl/Bayesian%20decoding%20of%20brain%20images.pdf
and the SPM implementation of it, but I've not actually tried it yet.
So far the decoding that I've done has been based on my own home-made
scripts (I find that's the only way that I can actually understand
what's going on), using classifiers or similarity-based approaches.
That multivariate Bayes paper looks interesting, although I confess
that it's somewhat opaque to me.
I'm currently exploring Mitchell/Gallant-style regression-based
approaches, which are basically just GLMs with parametrically
modulated regressors and lots of columns, hence my interest in seeing
whether standard SPM can be used as-is to efficiently implement them.
It looks to me like it probably can.
Thanks again for your help and feedback. Much appreciated!
Raj
|