Dear Louis,
> We are processing data from the 1st continuous EEG/fMRI experiment. We
> have spikes occuring at random and scanning taking place continuously
> over 30 minutes (total: 1200 volumes; 20 slices/vol). We want to
> investigate the interactions between successive spikes and want to use
> the Voltera formalism. We have idenitfied 4 types of spikes. Up to now,
> we have used Fourier expansions to do an event-related analysis - and
> found good activations.
>
> When we try to select the Voltera option in SPM99, the computer
> basically crashes: the amount of RAM used inflates to 1GB and nothing
> happens for a long, long time. Our best system has 512MB RAM; it seems
> that Voltera is RAM hungry - true? Any suggestions? Should we use an
> HRF instead of the Fourier expansion?
Your diagnosis is correct. The number of regressors in a second-order
Volterra expansion of the stimulus functions increases [roughly] with
the square of the number of basis functions times the square of the
number of trial types.
Our MRM paper, using this approach, used 6 basis functions (but only
one trial type). If you have four trial types (i.e. spike classes) I
would use the canonical HRF and its deivative (i.e. 2 basis
functions). Using one basis function (i.e. a canonical HRF) is
equivalent to a conventional first order model that includes explicit
interactions among trials.
With best wishes - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|