> 2.There is a recent paper in Neuroimage(http://www.sciencedirect.com/science/article/pii/S1053811911011815) suggesting that the movement would have a substantial influence on resting state results.The authors suggested to use derivatives of a single measure of intensity on each volume as an index to identify severe movement.I am just wondering what is the best way of removing movement from resting state data? I can see in the graphic interface that MCFLIRT is used to realighn the data, but is this really sufficient? Is there any easy way to identify a component(s) which reflects movement and remove it from the data?I will appreciate if someone can refer to any tutorial of doing this?
You can regress out the six rigid-body motion parameters generated by mcflirt prior to doing ICA. I think fsl_glm can be coerced into doing this. You can also discard individual volumes with high motion. Our lab censors on the basis of outlying costs, i.e. the value of the cost function between each motion-corrected volume and the reference volume. You can use a combination of mcflirt, fslsplit, and flirt to get the cost at each timepoint. In my limited experience, doing all this does not make a "substantial" difference when using ICA, since motion artifact tends to precipitate out in its own component anyway. The paper to which you are referring deals with seed-based voxel correlation, for which it is absolutely correct that motion makes a difference.
|