Hi Rick;
Well, this whole issue is arising during translation back and forth between AFNI and FIASCO (www.stat.cmu.edu/~fiasco). Both systems use a header file with key-value pairs, and the two sets of pairs are similar, but the two systems use different coordinate systems. FIASCO tracks the spatial position of the 8 corners of the data volume while AFNI tracks 1 corner and uses DELTA to infer the other corners.
This is all sensible enough. We've recently begun using 'adwarp' to Talairach datasets, though, and we notice that its transformation flips the images around in the cube. For example, suppose I take the file axial+orig and use AFNI to Talariach it. if axial+orig.HEAD has ORIGIN { -118.4355,99.34125,92.944}, DELTA { 0.9375,-0.9375,-1.5}, and ORIENT_SPECIFIC {0,3,5}, then axial+tlrc.HEAD has ORIGIN {-80,-80,-65}, DELTA {1,1,1}, and ORIENT_SPECIFIC {0,3,4} . You'll notice that something is fishy already, because the second component of DELTA has changed sign but the corresponding component of ORIENT_SPECIFIC is still 3. The layout of the anatomy in the file has also changed- the top of the head was originally at the beginning end of the +orig.BRIK file and is at the trailing end in the +tlrc.BRIK file. I believe the left and right sides are also swapped.
So what's up with the swapping of the data within the BRIK, and how is it reconciled with the changes in DELTA, ORIGIN, and ORIENT_SPECIFIC? I'd like to figure out what voxels correspond to what locations in physical space without inspecting the BRIK images, but I can't infer the rules to get that from the quantities in the .HEAD file.

Thanks very much,
-Joel