So, it looks like the motion censoring is quite severe in this case.
Looking at 1d_tool.py, it takes the motion correction parameters (3 rotations in degrees + 3 translations in mm), finds the forward differences then computes the root-sum-of-squares over all six parameters. Then, for any TR that survives a threshold of 0.2 it means that (1) the "displacement" of the rotation parameters was less than 0.2 degrees and (2) the displacement of the position parameters was less than 0.2 mm (for TRs that don't survive, the mixing of rotations and displacements is difficult to interpret).
Thinking only about the position part (2), should we think of these thresholds as global truths--"subjects shouldn't move more than 0.2 mm per TR regardless of the acquisition resolution" or should we pick the threshold relative to the raw data resolution?--"subjects shouldn't move more than 10% of the acquired resolution per TR?". I'm inclined more towards the later because that's what would be relevant on the interpolation grid. Possibly movement specifically in the slice-selection direction could be more important for censoring due to relaxation effects.
What resolution data does the 0.2 advice come from?
Edited 4 time(s). Last edit at 08/09/2012 11:14AM by judd storrs.