External Email - Use Caution
Hi! (tl:dr - I have a few questions listed at the end of this message. most of the message is just details about these questions).
I have a volumetric map of values for each subject, which you can think of as a contrast (it is actually a contrast of distances from a Representational Similarity Analysis (RSA) calculated in Matlab using the RSA toolbox, but that’s irrelevant to my questions). I did my preprocessing in FSL. These maps are currently in a downsampled (2x2x2) version of the anatomical image (see next paragraph for details). I used FSL’s FLIRT to move my functional images to that space, using the examplefunc2highres.mat transformation matrix, highres (the anatomical image) as the reference image, and the flag -applyisoxfm 2 to ensure the resolution stays 2x2x2.
(( **Less relevant to my Freesurfer questions so feel free to skip**: the reason I chose to run my RSA analyses in this 2x2x2 anatomical space is that I have 8 runs (EPI sequences), with a structural scan after 4 runs. Since I need to align all my runs to one template image to do the RSA analysis with conditions across runs, it seemed like a natural choice: I first run FEAT preprocessing on each run separately, and use a the middle image from run X as the motion correction template for run X (as is the default). I then transform the filtered_func_data.nii.gz file of each run to the 2x2x2 anatomical space. If this seems like a mistake to you, please let me know! ))
I would like to project these contrast maps to a common surface, and to do group analysis on the surface. I think I need to use the command mris_preproc or mri_vol2surf (see next paragraph), but both need a registration file which I’m not sure how to get. given that my contrast images are already in the “correct” (anatomical space) orientation, I thought the transformation matrix should just be the identity matrix. I tried using bbregister with the brain extracted structural image as the --mov template (the resolution in the resulting register.dat file will be 1x1x1, and not 2x2x2 like the transformed functional images, but the resolution lines in the register.dat file are anyway being ignored. The transformation matrix should be correct). I tried this and the transformation matrix was not the identity matrix, but rather: 9.999998211860657e-01 5.568005726672709e-04 -6.800168921472505e-05 5.548541247844696e-02 1.512142625870183e-04 -1.508450955152512e-01 9.885573387145996e-01 -3.581156954169273e-02 5.401717498898506e-04 -9.885571599006653e-01 -1.508451700210571e-01 3.751721233129501e-02 when I look at this registration with tkregisterfv it looks fine (both for the structural image and for the examplefunc after transformed to the 2x2x2 structural space) - so this seems to be working. If anybody has an explanation why the transformation matrix between the skull-striped anatomical image from FSL to orig.mgz is not the identity matrix I would appreciate it.
I’m not sure how to get the maps into a common surface and not subject surface. The register.dat file I will obtain will be the registration from the functional space (in my case the downsampled anatomical space) to the subject’s anatomical space. If I give that as input to mri_vol2surf, but use "--trgsubject fsaverage", will Freesurfer automatically know to apply the registration from the subject’s surface to fsaverage (if I'm correct it uses the sphere registration - is that the right thing to do?)? If I only want to do surface group analysis and I’m not interested in viewing the individual subject’s contrasts on the surface, do I even need to use mri_vol2surf? or is it enough to just use mris_preproc and use the --iv flags to specify the volumetric contrast images and the register.dat files?
I would like to run a one-sided Wilcoxon sign-rank test on these contrasts across subjects (and not a GLM - I want to know if my contrasts are different from 0 across subjects, but the contrast maps are not necessarily normally distributed) . so I can’t use mri_glmfit. Is there any way of doing that in Freesurfer? If not, is there an easy way to export the output of mris_preproc (stacked contrasts of different subjects, already sampled to fsaverage) into Matlab? After doing the analysis in Matlab, can i just write the result (a vector of p-values of size nNodes x 1) into a csv file or something like that, and then load it in Freeview as an overlay on the fsaverage surface? Is there a particular way in which this conversion from freesurfer to Matlab and back to freesurfer should be done?
Finally, I will need to correct for multiple comparisons. Is there any way for me to still do that with the Freesurfer machinery for cluster correction, even though I am not running mri_glmfit, but rather doing a non-parametric statistical test?
To summarise, my questions are:
1) How to create a registration.dat file with registration from my weird 2x2x2 anatomical space, where my (functional) contrasts currently are, to the anatomical space (same orientation, different resolution). 2) How to use those registration.dat files correctly to prepare the contrasts for group analysis - do I need to use mri_vol2surf or should I use mris_preproc? 3) Is there a way of running a one-sided Wilcoxon sign-rank test in Freesurfer? 4) What is the best way to export .mgh files to Matlab, and save them from Matlab in formats that are Freesurfer-compatible? 5) If I am not running a GLM, is it possible to still use Freesurfer for multiple comparisons? If so, how?
Thanks very much for any help!