The effects of motion on parametric fMRI analysis techniques

Subject motion during the time course of functional activation studies has been shown to cause spurious signals which can mimic 'true' activation. Therefore, the importance of motion correction has been widely recognized. Correction with post-processing using image registration software is common practice in functional imaging and analysis. Many image registration algorithms, developed for analysis requirements other than fMRI, assume rigid body motion. Although these techniques are now routinely used by a number of groups, rigid body coregistration has not yet been shown to reduce the effects of motion to an acceptable level in fMRI analysis, i.e. the effects on resulting correlation analysis directly. In this paper we have used volume data to assess rigid body co-registration in terms of motion artefacts for the different correlation approaches used in fMRI. We have developed a new way of visualizing motion effects in correlation analysis based on generating a scatter plot of correlation score against local image gradient. This technique has been tested on fMRI data sets from a functional paradigm suffering from motion correlated artefacts, with and without rigid body motion correction. Although we do not attempt to estimate the actual residual motion, this technique can be used to verify the results of analysis and select regions of relatively unambiguous activation. This paper assesses directly the rigid body assumption and proves the need for, and effectiveness of, co-registration for all correlation based analysis techniques. The specific differences between the popular correlation forms used are investigated and explained. We show that for certain forms of correlation analysis the effects of motion, while not removed altogether, are effectively statistically eliminated.