Hi,
I'm using FSL's mcflirt for motion correction with the "-mats" option to store a registration matrix for each time point. In my application, I would like to use a one-step interpolation process to sample points from the freesurfer surfaces (e.g. white) from the 4D fMRI data.
I tried to convert the FSL .mat matrix to "TkReg-RAS space", which, as I understand it, is the space the surface coordinates in "lh.white" etc. are stored in. What I did:
1) Obtain the .mat matrix file using mcflirt:
mcflirt -in bold.nii.gz -out bold_mcflirt.nii.gz \ -reffile bold_ref.nii.gz -mats
this creates a directory "bold_mcflirt.nii.gz.mat", where the registration matrix files for each time point are stored (MAT_0000, MAT_0001...). As I understand it, the FSL matrix is the transform from bold to bold_ref in mm space. I.e., to get the transform in voxel coordinates, one would use:
M_vox = A1 * M * inv(A2)
where A1 and A2 are the affine matrices from the headers in bold_ref and bold, resp. The matrix M is the affine matrix computed by mcflirt.
2) Convert the matrix to TkReg-RAS space. The matrix bold->bold_ref matrix in this space is given by
M_ras = vox2ras1 * M_vox * inv(vox2ras2)
where vox2ras1 and vox2ras2 are obtained using "mri_info --vox2ras-tkr" for bold_ref and bold, resp.
Is this correct? I have tried it and for some reason the translated points don't seem to be at the correct locations in the "bold space". What I did is create an all-zero nifti file with the same size as "bold" and then fill-in the points corresponding to the white surface. For later time points (when the subject movement relative to bold_ref is larger), the surface is increasingly misaligned. This tells me that either
a) the transform is computed incorrectly
or
b) MCFLIRT stores something different than the bold -> bold_ref transform for each time point.
Any help is appreciated.
Best,
Martin