Dear all
a recent paper (Murphy and Garavan, Neuroimage 2005 Oct 1;27(4):771-7) proposes a method of estimating activation that might be more stable with respect to the standard approach. In synthesis, an impulse response function (IRF) is estimated with 3dDeconvolve, then a non-linear fitting of a gamma function is applied voxel-wise to the estimated IRFs, and finally the area under the curve of the fitted gamma function, normalized as percent change with respect to the baseline (AUC%), is computed. This parameter, instead of the usual estimated *amplitude* of the fitted hemodynamic response function, is then entered into a group-level statistics (t-test or anova) to assess between-subject effects.
In trying out the approach, I was wondering about the optimal values for the initialization parameters of the nonlinear fitting of a gamma function, considering that the experimental design is a standard event-related task with brief event durations (~1 sec). Since the expected hemodynamic response shouldn't be *very* different from the canonical one (crf. Cohen), it should be possible to have an educated guess about the initialization values, which would maximize the performance of the algorithm in terms of both computing time and of the goodness of fitting. Also, it is not clear to me what would be a sensible choice for the 3dNLfim parameters nrand and nbest.
thanks for any comment or suggestion
giuseppe