That is not scanner vibrations, but rather is probably the effects of eddy
currents. Subject motion would tend to be a general trend not a regular
back and forth, and vibrations cause a different artifact (it is too fast to
show up as bulk motion of the volumes, rather it shows up as signal loss,
especially in the medial superior parietal region). The steps between each
average are probably from slight differences in the central frequency
leading to bulk translations of the whole timeseries.
1) I don't know if there is a clear threshold for this. When you start
seeing motion artifacts in the raw images the data is probably not useable.
2) Translations are the 4th column of the matrix
3) I would use avscale for this purpose. Use the --allparams flag to get
the angular rotations in radians.
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf
Of Alina Jurcoane
Sent: Wednesday, January 20, 2010 10:52 AM
To: [log in to unmask]
Subject: [FSL] eddy_correct
To the DTI people out there,
we are analyzing dti data with three repetitions of the same dti sequence:
dti1, dti2 and dti3.nii.gz. As previously discussed on the list we first do
fslmerge -t dti_merged dti1 dti2 dti3 followed by eddy_correct dti_merged
After eddy_correct I plotted the x,y,z translations (4th matrix column in
the ecclog) and got graphs as in attachement (and that is a nice graph, some
patients had motion of 3-4 up to 5mm). It is obvious that there is little
motion during the b0 images (timepoints 0-9, 70-79, 140-149) while, during
the gradient weighted images there is a continous back and forth motion
My questions are:
1. What would one consider "too much motion" to lead to exclusion of a
particular subject or dti run from further analysis?
2. Did I corectly identify translations as the fourth column in the
3. How would one compute rotations (I want to make a plot for translations
and one for rotations, as in mcflirt)? Is it through avscale or how?
Thanks for the help,