 |
 |
Hi, Introducing regressors from MCFLIRT into the GLM and using confound EVs from fsl_motion_outliers work in very different ways. The first case assumes that there is a linear relationship between the observed signal and the detected motion with MCFLIRT. The second is used to identify timepoints where the residual image intensity changes are over a threshold (and hence outliers) and then "removes" the effect of these timepoints completely from the data. It is different from editing out the timepoints as it does not cause the problems for filtering and prewhitening that editing does. Therefore it you have very strong motion at a point then fsl_motion_outliers should, in theory, work better as the effects on the image intensity are using very non-linear and so the MCFLIRT method normally does not work so well here. However combining them is a potentially good, if conservative, thing to do. I certainly expect different results. Also, I am unaware of anything called fsl_motion_remover so I cannot comment on this. Is it something developed in your own lab? As for MELODIC - this is a powerful way of removing motion effects if you can clearly identify them in the output. As it has a subjective component, it can be tricky in some situations to be sure which components are definitely motion effects, but often a combination of where the effects are spatially (normally around big intensity boundaries) and when they are in time (often sharp spikes) allows you to work out which components are the ones you need to remove. It is also OK to remove some but not all in case you are unsure - it will then just be a question of whether you have improved the situation sufficiently to see the true biological activations come through. Just try it and see how it goes. Hope this helps. All the best, Mark On 16 Feb 2010, at 16:36, Brett Harton wrote: > Dear FSL Pros, > > I am fairly new to FSL and I'm still trying to figure out all the > options available > to produce optimal results. > > My subject data is on severly brain injured subjects in whom it is > difficult to > control movement during fMRI. > > In FEAT, I am using MCFLIRT, and I also tried adding the output from > fsl_motion_outliers as a Confound EV. I got different results when > using this > Confound EV than when using MCFLIRT alone. > > My question(s) is/are: What is the difference between > fsl_motion_outliers and > fsl_motion_remover and should I use one over the other to correct > motion? > Or, should I use both and in what sequence? Finally, would MELODIC > be a > better solution for my motion dilemma? > > I hope my questions are clear. Thank you for your help! > > -Brett Harton >