Hi Ziad/Peggy,
I empirically tested this idea out (substituting the RMS error column data from -rprefix as an in-plane motion vector). I created from the raw slice data two additional time series datasets. In one dataset, 2dImReg was applied before 3dtshift, (since it acts slice-wise) and in the other dataset, 2dImReg was run on the interpolated time-series volumes after 3dtshift.
Running 3dDeconvolve with both alternate datasets indicated that I had lost some power compared to using just 3dvolreg with its -dfile output in the model. This suggests that in-plane motion was not sufficiently controlled for in the model when I tried to approximate it with .oldrms output.
I suspect that RMS would be higher in volumes when motion has occurred, but it's not the same datum as an actual calculated translocation, as is obtained when using -dfile in 3dvolreg. The closest approximation in the existent 2dImReg is the output from -dprefix, where one uses the -dmm operator so as to get that output in mm.
Unfortunately, that .dx .dy output is separately calculated and dumped out for every image collected. In my case, since the NMRF3t lets you obtain 16 slices per second, that's 9600 lines of data for a 600 second, 300 volume scanning run!
Is there any way 2dImReg could be explanded to allow for calculation and dumping of a *volume-wide approximation* of in-plane adjustment across the whole volume, so you get a text output that equals the number of volumes to easily put into a model? Alternatively, could an operator be added to the program that would at least dump out corrections for only a single slice of the montage, such as the base slice as a default?
Jim B