Right now I'm not interested in the eddy currents. I have DTI data corrected for eddy currents and head motion, and would like to add simulated motion to test how well different algorithms can correct for the motion.

And will this “added” motion then be considered ground truth? That would of course hinge crucially on the initial correction being absolutely perfect, which makes it all a little circular.

I would recommend using the simulations by Graham, Drobnjak and Zhang as your ground truth instead. It is my understanding that they are planning to release simulated diffusion data along with ground truth. If any of them are reading this they might be able to confirm?

Compared to fMRI data, I think that motion correction of DTI data is more challenging, since DTI volumes have different intensity and noise levels (depending on measurement direction and b-factor). I would therefore like to only use the motion correction part of eddy or eddy_correct, but maybe that's impossible.

As long as your initial manipulations hasn’t cause areas of NaN (or zeros) at the edges of the FOV it should be fine with either eddy or eddy_correct.

My fMRI software BROCCOLI uses an image registration algorithm that is not based on image intensity, instead it uses local phase from quadrature filters (and the local phase is invariant to the intensity level). My hypothesis is that this algorithm may be better for images of different contrast/intensity/noise levels, compared to methods based on intensity.

Good luck Jesper

- Anders

2015-09-21 11:10 GMT+02:00 Jesper Andersson <[log in to unmask]>:

Dear Anders,

> what is the most common way to correct for head motion in DTI data? To use the function eddy_correct?
>
> If I only want to do motion correction and no eddy correction (assuming it has already been done using some other function), should I then use flirt? mcflirt?
>
> Has there been any comparison of DTI motion correction tools, especially for high b-values?

there is a paper coming out soon (I think) from the UCL (Mark Graham el al.) group comparing eddy_correct and eddy. They also present a simulation framework that will hopefully facilitate more comparisons of eddy current/subject motion correction methods in the future.

I don’t want to reveal too much of their findings prematurely, but they find that eddy_correct performs worse than eddy. The strategy of eddy_correct, i.e. to use an affine transform and a cost-function that is “more accepting of differences in contrast”, is similar to some other methods of motion correction of diffusion data so it is likely that their conclusions extend to some of those. Though it remains to be proven of course.

My recommendation would be to use eddy if your data is such that it allows for it. In principle it should be fine with data that has already been corrected for eddy currents, but it depends on what that method does to the edges of the FOV. My personal preference would be to just use eddy to correct for both effects. Just out of interest, what method are you using for the eddy currents.

Good luck Jesper

--

Anders Eklund, PhD