Hi Prad,

All "scan-to-scan motion" measures, including Power's frame displacement, Jenkinsons's rms displacement, Yang's average voxel displacement, etc. as well of course as ART's composite motion measure, produce similar but unfortunately not identical measures (i.e. threshold values for all of these measures are only roughly comparable). Details differ, likely because there is probably not clearly "best" way to define a single motion index that summarizes the effect of a rigid body translation/rotation on the brain volume. Because of this, if you really need to compare them my recommendation would be to actually compute those measures explicitly in your data (many of these measures can be directly computed only from the realignment parameters -the rp_*.txt files produced during realignment).

In practice, recommended conservative threshold values for these measures often will consistently produce something around 5% of data removed by censoring in a healthy adult population with low movement. In CONN, the "conservative" settings, for example, have been defined exactly in this way to correspond to approximately 5% of data removed in our control normative population sample, so that would be a reasonable conservative choice (cf. "liberal" settings in CONN correspond to approximately 1% of data removed in the same sample). Alternatively you may find the threshold values that work best in your sample, and/or find your own optimal thresholds using for example a control group. In general, beyond the specifics of the method/threshold used, I find it is always a good idea to: a) report, in addition to the specific thresholds used, the percentage of data being removed by censoring/scrubbing, since that is one simple and informative measure that addresses the extent of censoring being performed beyond the specificities of the scan-to-scan motion and BOLD signal measures used; and b) compute subject-level summary measures of motion/censoring levels and use those as covariates in your second-level analysis, in order to convince yourself and the reviewers/readers that your results are not affected by residual differences in motion between conditions/subjects/etc. 

Hope this helps
Alfonso
 

Originally posted by prady umna: