23
12 s. Event-related data were analyzed using the finite-impulse response (FIR) method,
as implemented in the SdDeconvoIve program in the AFNI software package (Cox 1996).
Tent shaped regressors were created for each of the nine images (18 s window with a TR
of 2 s) after trial onset. Because a rapid event-related design was used, at any point in
the MR time series the image intensity contained contributions from the overlapping
responses evoked by many previous stimuli. By assuming linearity and time invariance,
the FIR method deconvolved the overlapping responses into separate responses, one for
each stimulus type, that were equivalent to those that would be measured with a slow
event-related design.
In the general linear model, regressors of no interest consisted of the motion
estimates from volume registration and polynomial regressors to account for baseline
shifts and linear drifts in each scan series. The ratio of variance accounted for by the
stimulus regressors and regressors of no interest to the regressors of no interest alone
was used to calculate an F-ratio and associated significance for each voxel.
Corticalsurfacemodels
Individual cortical surface models were created using FreeSurfer software (Fischl,
et al. 1999). Surfaces were visualized and average cortical surface models created using
SUMA software (Argali, et al. 2006). To visualize activity buried in sulcal depths, the
surface was partially inflated using 500 iterations of a smoothing algorithm. Anatomical
features on the partially inflated surface were visualized by colorizing the surface with a