There are a lot of obscure AFNI programs written for specific purposes, for specific people; some of these programs might be useful to a wider audience. So I'm going to highlight one such program weekly (if I have the discipline to keep this up).
First such program: 3dTcorrMap
This program is for exploring resting state data, and computes some measure of how 'connected' each voxel is to the collectivity of all other voxels.
First, it computes the correlation coefficient of every voxel 'x' in a time series dataset with every other voxel 'y': rho(x,y). Saving all these numbers would take up a LOT of space. Instead, 3dTcorrMap computes for each voxel 'x' some summary statistic(s) about the collection of rho(x,y) values; for example:
* the average over y of rho(x,y)
* the number of y's with |rho(x,y)| >= threshold
* the RMS value of rho(x,y), and so on.
The output is a dataset that gives this summary of the 'connectedness' of x to all other voxels in the mask.
3dTcorrMap was developed for use by a researcher in the NIMH, but has also been used by at least one other person to make a poster at the just-concluded HBM-2009 meeting. So I decided that more people might like to be aware of this utility.
3dTcorrMap is kind of slow (10s to 100s of minutes), and does benefit from OpenMP parallelization (added a few weeks ago). Only the Mac OS X 10.5 versions of AFNI are precompiled with OpenMP enabled, but if you have GCC 4.2 (or above) on your Linux system, you can manually compile the AFNI programs yourself with OpenMP -- the programs that have been modified to use OpenMP will run significantly faster.