Dear Hamed,
>
> Dear Jesper
> I guess there is a mistake in the text. Is this what you meant?
>
> average the 2nd column = a measure of “short term movement”
> average the 1st column = a measure of “long term movement”
yes, you are of course right. Sorry for that.
> And may I ask what is the common threshold for these 2 measurements to exclude a subject?
I don’t think there is a commonly accepted threshold, nor am I sure there should be one. As we, and other groups, become better at modelling and correcting for more of the adverse effects of movement we should hopefully be able to use data with more and more subject movement.
I also think that it will, and should, depend on the actual project. The concern about movement is that “uncorrected” movement effects will affect the response variables we are working with (for example FA). So if we are for example working with a group of young healthy motivated subjects it might be that even “a moderate” amount of movement in a given subject might render that subject an outlier in terms of FA, and maybe that subject should be discarded.
In contrast, if you are working with a group of babies all subjects will move much more. That will then affect FA, and will mean that we lose some statistical power. But for a subject to be considered an outlier, motivating its removal, the movement would have to be “very large”.
In the, hopefully not too far away, future we plan to release tools that will take the output from eddy, run statistics on those and generate reports that highlights subjects as outliers in the particular project in question. We hope that will be helpful in compiling the information in a format that is more easily accessible. But at the end it will always be a “common sense” call which subjects to discard and which to keep.
Jesper
>
> Thanks and best,
> Hamed
>
>
> Dear Jan,
>
> I would recommend using the “eddy_restricted_movement_rms” info. If you average the second column it would give you a measure of “short term movement” which would cause variance due to things like slice-to-volume movement and spin history effects. If you average the second column it would give you a measure of “long term movement” which would cause variance due to for example susceptibility-by-movement interaction and “movement within a stationary bias field”. You could then use those two as covariates in your model.
>
> An example of a paper looking at things like this would be Anastasia’s paper (http://dx.doi.org/10.1016/j.neuroimage.2013.11.027 )
>
> Jesper
>
|