We've done a little realtime fMRI using AFNI with the goal of guiding MRS voxel selection on our Varian 4T. Our reconstruction is in house IDL code and the images are fed to the plugin using IDL's sockets.
The unanticipated complication was not being able to send individual volumes to the plugin one at a time (unless I missed how to do that). Keeping the iochan alive between volumes, means that either everything has to be driven by IDL, or we have to implement some other communication protocol with an intermediate program. What I'm trying to say is that it was a more overhead to maintain the iochan state between volumes, etc and in retrospect I don't really see why it was necessary. It could have been much easier if rtfeedme had a mode to simply append volumes to the end of the growing realtime imaging dataset. As a bonus, if rtfeedme did that, then it would be easier to experiment with data preprocessing scripts.
--judd