Dear Flora,
as structural T1-weighted images are acquired as 3D volume in general, which takes several minutes till a single output file is saved, you don't know about head motion. You could run some kind of localizer right before and directly after the structural volume, and then estimate the movement between these two time points, but this doesn't help much because you won't know whether the subject moved around during the structural volume.
So in the first step you or someone else should definitely check the quality of the data manually for every single subject. If you're familiar with motion artefacts then it's quite easy to detect them qualitatively. Still, it might be better to use a quantitative threshold in addition, as there might be other problems (low SNR due to shimming errors, ...). One option would be signal-to-noise ratio.
In a second step you could go with a feature "Check data quality" in the VBM8 toolbox by Christian Gaser. This way you can load many structural volumes (or the segmented GM files) at the same time and plot covariance matrices for all the volumes and boxplots to detect outliers.
Best,
Helmut
|