The program is based on the paper by Ding, the first reference listed in the help for 3danisosmooth. The program can indeed be used for image datasets other than DWI datasets. The help is available, as for most AFNI programs, with the -help option (or the -h_view option). The help has not changed since its inception, so you may have seen all that is available. Briefly, the deltat option represents the simulated pseudo-time, a step-size, for each iteration of the simulation. By specifying smaller step sizes, you can more finely tune the smoothing at the cost of performance speed. Our version uses larger steps than those used by Ding in their original paper although I believe they have since followed with similar larger steps as used in 3danisosmooth. The edgefraction corresponds to the fraction of "edginess" compared to the fraction of gaussian smoothing applied. The default is 0.5, but it can be increased up to 1.0 or down to 0.0, in which case, the program becomes a very inefficient way to apply a gaussian smoothing operation.