This may seem out of step with the messages typically on this board, but seems like the best avenue. Some of the studies we have been analysing are taking tens of hours to process. We are also getting to the point where we would like to process scans in realtime (or very close to it) and feed the results (after some additional postprocessing) back to the scanner/TMS positioner ... as parameters for positioning, pulse parametrics, etc. To do this we've been looking at processing the data with a Beowulf cluster. A standard programming interface (MPI) could be employed to "parallelize" AFNI to a wide variety or parallel architectures. Is there any work underway to do this? Is there general interest in this? (Lest you panic - what I have in mind is doing this work here with a collection of folks from the Univ. of South Carolina, Medical Univ. of South Carolina, and the South Carolina Research Authority) and working with the AFNI owners to mainstream it. Comments?
Dave Ramsey