Print

Print
Hi Erin,

Please see below:

On 8 July 2016 at 16:27, Erin Drazich <[log in to unmask]> wrote:
Thanks, I think I'm getting there! Okay, so for the full analysis after motion correction, do I plug in the output of the preprocessing feat (ie, reg/filtered_func_data), or do I got back to my original bold file?

I would start cleanly, only adding the nuisance EVs (produced with the earlier run) to the model.

 
Can I run the full feat on images in standard space (with input image reg_standard/filtered_func_data from featregapply)? Do I need to make any corrections to do that? Or just make sure all my inputs are in standard space? I've seen a few people suggesting that, so I wanted to clarify.

Just start cleanly and follow the manual. It will prevent unexpected behaviour later.

 
You had also mentioned using for the T1 in standard space from feat. Where is this image found? I haven't found any images in the feat directories that look like the T1 in standard space if I run flirt directly on the T1.

In the first run (the one to obtain the nuisance), select "Preprocessing" in the pull-down menu, enter the functionals in the "Data" tab, and the relevant structurals in the "Registration" tab. Disable the filters and run. Once it finishes, check the directory "reg" inside the output directory. There will be various files in different spaces. It should have one that matches what you need, if not, it also contains other files that can be used with FLIRT and FNIRT for the alignment. Use the T1 aligned to the functional space as input to FAST.

 

Finally, for the exchangeability blocks, I just wanted to check the glm I've devised:

Group   Subject EV PT   EV CTRL
1       Pt1_run1        1       0
1       Pt1_run2        1       0
1       Pt1_run3        1       0
1       Pt1_run4        1       0
2       Pt2_run1        1       0
3       Pt3_run1        1       0
3       Pt3_run2        1       0
3       Pt3_run3        1       0
3       Pt3_run4        1       0
4       Pt4_run1        1       0
4       Pt4_run2        1       0
5       Ctrl1_run1      0       1
5       Ctrl1_run2      0       1
5       Ctrl1_run3      0       1
6       Ctrl2_run1      0       1
7       Ctrl3_run1      0       1
7       Ctrl3_run2      0       1
8       Ctrl4_run1      0       1
9       Ctrl5_run1      0       1
9       Ctrl5_run2      0       1

Do I need to use PALM or can I still do dual_regression and randomise? All the runs are equivalent and obtained on the same day, so there are no paired repeated measures.

PALM can be used, but in this case it's necessary to make a strong assumption of compound symmetry within subject, and also use multi-level exchangeability blocks, so that subjects with same number of runs can be shuffled with each other (e.g., Pt4 and Ctrl5). Another option is Bryan Guillaume's SwE toolbox: http://www2.warwick.ac.uk/fac/sci/statistics/staff/academic-research/nichols/software/swe

All the best,

Anderson
 

Thanks again so much! Happy Friday!