 |
 |
Hi, I would recommend the following: - run the first level analysis with FEAT for all subjects and have everything turned on, including registration, but deselect the standard image and only select the Main structural image (this will generate the example_func to highres registrations) - re-run one case with just registration turned on (everything else off) and select both structural and standard images, including the non-linear registration (this will generate the desired highres2standard registrations) - probably best to call the output something quite different here to avoid confusion with the previous feat directories - copy all the highres2standard files from the session above to all of the .feat/reg directories of the sessions for that subject - in each session's feat directory run updatefeatreg This will then generate full first-level analyses with all the registration details done, but only having run the highres to standard registration once. Regarding the motion EVs, you can simply select this as normal in the first step above. For the fsl_motion_outliers, I'm afraid we are still in the process of running experiments and preparing a paper so I can't really comment on relative performance. Anecdotally it seems to work well on some datasets but not all. It does use a different way of determining the outliers (for example, if motion correction fails for a volume this will be detected here but may not be obvious from the mcflirt rms file, and it is also more sensitive to cases of strong within-volume motion effects). To use it you just need to run this initially, create appropriately named confound ev files and then specify them in the FEAT setup in the first step of the above analysis. Hope this helps. All the best, Mark On 4 Jul 2009, at 15:15, Jonathan Ipser wrote: > Hello everyone, > > I was hoping for some reassurance that I am on the right track re > 1st level > analyses. > > The data I have consists of a total of 24 runs per participant (12 per > session). Rather than running the entire first-level analysis using > the FEAT > GUI for each of these runs, which would have involved registering the > structural to standard images 24 times per participant, I attempted > to save > some time by conducting the structural to standard registration only > once > per subject per session, using flirt via the command line. I have > still used > FEAT for motion correction, but made sure I deselected all the > options on > the registration tab. > > I would now like to use FEAT for the remainder of the 1st level > analysis. I > am assuming that the 4D image I should use as input for the stats > step is > the filtered_func_data.nii image in the main output directory. Is > this correct? > > I would also like to add motion parameters as events of no interest > using > FEAT. However, the "add motion parameters to the model" button is > disabled > unless you conduct the prestats and stats in the same FEAT run. From > reading > this mailing list, it seems that one can add the 3 translation and 3 > rotation parameter estimates from the prefiltered_func_data_mcf.par > file > using the "add additional confound EVs" button. Using the 1-column > format, > it should be possible to select this file directly, as the format of > the > prefiltered_func_data_mcf.par is consistent with the format of > multiple > 1-column EVs in the same file (TRs = rows, and EVs = columns). Is > this the case? > > Finally, although I intended including columns modeling individual > spikes in > the timeseries (identified from the > prefiltered_func_data_mcf_rel.rms file) > as additional EVs in this confound EV file, it seems that a better > option > might be to use the fsl_motion_outliers script to generate these EVs > (which > then can presumably be added to the confound EV file). Is there any > literature on the fsl_motion_outliers algorithm, perhaps comparing the > performance of these two methods? > > Best, > Jonathan Ipser > Department of Psychiatry and Mental Health > University of Cape Town >