On Fri, 9 Jul 2004 13:05:34 +0100, Joseph Devlin <[log in to unmask]>
wrote:
>Dear Conny,
>
>The raw parameter estimates are really in arbitrary units with a scaling
>depending on both your design matrix and the values of the data you were
>modeling. So to get % BOLD signal change you need to take both factors
>into account. If you multiply the PE value by the difference between the
>max and min height of the EV (after convolution, if present) and then
>divide the result by the mean value of the time series in that
>voxel/region, you end up with what you want.
Is this correct for all designs? I thought that one would want to multiply
by the (max-min) of the ideal HRF *before* convolving with the stick
function that represents the train of events/stimuli. The reason being that
if HRFs overlap, as in a rapid event related design, the (max-min) of the
EV after convolution will contain values that represent the addition of two
(or more) HRFs.
Tom Johnstone
University of Wisconsin-Madison
|