Adapting for fMRIPrep; Surfaces Do Not Load (sphere mismatch)

[smeisler]

Hello,

I am adapting the HCP tutorial for fmriprep outputs. I generated CIFTI outputs with fmriprep's --cifti-output 91k argument. This dataset has 3 runs of data. Thus, fnames_ts, events_all, and confounds_all are size-3 lists, as described in the documentation. I use the following gifti file (left hemisphere example): /PATH/TO/fmriprep/SUBJECT/ses-1/anat/SUBEJCT_ses-1_hemi-L_midthickness.surf.gii. I am running into an error described below:

Command

results <- BayesGLM_cifti(
    cifti_fname = fnames_ts,
    surfL_fname = fname_gifti_left,
    surfR_fname = fname_gifti_right,
    brainstructures = c("left", "right"),
    onsets = events_all,
    TR = TR,
    nuisance = confounds_all,
    nuisance_include = c("drift", "dHRF"),
    scale_BOLD = TRUE,
    scale_design = TRUE,
    GLM_method = "both",
    ar_order = 6,
    ar_smooth = 5,
    session_names = c('run-1','run-2','run-3'),
    resamp_res = 10000,
    num.threads =  parallel::detectCores() - 2,
    verbose = TRUE,
    outfile = NULL,
    return_INLA_result = TRUE,
    avg_sessions = TRUE,
    trim_INLA = TRUE,
    keep = FALSE,
)

Error

SETTING UP DATA 
    Reading in data for session 1
Error in read_surf(surfL, "left"): The object could not be converted into a surface.
Traceback:

1. BayesGLM_cifti(cifti_fname = fnames_ts, surfL_fname = fname_gifti_left, 
 .     surfR_fname = fname_gifti_right, brainstructures = c("left", 
 .         "right"), onsets = events_all, TR = TR, nuisance = confounds_all, 
 .     nuisance_include = c("drift", "dHRF"), scale_BOLD = TRUE, 
 .     scale_design = TRUE, GLM_method = "both", ar_order = 6, ar_smooth = 5, 
 .     session_names = c("run-1", "run-2", "run-3"), resamp_res = 10000, 
 .     num.threads = parallel::detectCores() - 2, verbose = TRUE, 
 .     outfile = NULL, return_INLA_result = TRUE, avg_sessions = TRUE, 
 .     trim_INLA = TRUE, keep = FALSE, )
2. read_cifti(cifti_fname[ss], surfL_fname = surfL_fname, surfR_fname = surfR_fname, 
 .     brainstructures = brainstructures, resamp_res = resamp_res)
3. read_cifti_separate(cifti_fname, surfL_fname = surfL_fname, surfR_fname = surfR_fname, 
 .     brainstructures = brainstructures, idx = idx, resamp_res = resamp_res, 
 .     mwall_values = mwall_values, verbose = verbose, ...)
4. add_surf(xifti, surfL = surfL_fname)
5. read_surf(surfL, "left")
6. stop("The object could not be converted into a surface.")

While running:
/rdma/vast-rdma/vast/gablab/smeisler/workbench/bin_rh_linux64/../exe_rh_linux64/wb_command -surface-resample /net/oms.ib.cluster/om4/group/gablab/data/ds003126/derivatives/fmriprep/sub-047EPKL011005/ses-1/anat/sub-047EPKL011005_ses-1_hemi-L_midthickness.surf.gii /tmp/Rtmp49kVVm/sphereL_32492.surf.gii /tmp/Rtmp49kVVm/sphereL_10000.surf.gii BARYCENTRIC /tmp/Rtmp49kVVm/resampled_10000_sub-047EPKL011005_ses-1_hemi-L_midthickness.surf.gii

ERROR: input surface has different number of nodes than input sphere

While running:
/rdma/vast-rdma/vast/gablab/smeisler/workbench/bin_rh_linux64/../exe_rh_linux64/wb_command -surface-resample /net/oms.ib.cluster/om4/group/gablab/data/ds003126/derivatives/fmriprep/sub-047EPKL011005/ses-1/anat/sub-047EPKL011005_ses-1_hemi-R_midthickness.surf.gii /tmp/Rtmp49kVVm/sphereR_32492.surf.gii /tmp/Rtmp49kVVm/sphereR_10000.surf.gii BARYCENTRIC /tmp/Rtmp49kVVm/resampled_10000_sub-047EPKL011005_ses-1_hemi-R_midthickness.surf.gii

ERROR: input surface has different number of nodes than input sphere

More info

The surfaces will load with load_surf so I am confident it is not a simple path issue. The current commands seems to have no option to input spheres and instead deafults to that in /tmp/Rtmp49kVVm/sphereL_32492.surf.gii. I have matching sphere files from the templateflow directory. How can I incorporate those?

p.s. I would be happy to share my example fmriprep script when done, in case it can add to your documentation. I imagine it may be of interest to many fmriprep users.

Thanks, Steven

[damondpham]

Hi @smeisler , it seems like the CIFTI resolution might not match the surface geometry resolution.

If you do summary(read_xifti(fnames_ts)), what is the printed output? This will show the numbers of data vertices on the left and right hemispheres.

Then, if you do summary(read_surf(fname_gifti_left)), what is the printed output? This will show the numbers of vertices on the left surface geometry.

[damondpham]

By the way, load_surf reads in the surfaces included in the ciftiTools package. To read in your own file, e.g. /PATH/TO/fmriprep/SUBJECT/ses-1/anat/SUBEJCT_ses-1_hemi-L_midthickness.surf.gii, read_surf should be used instead.

[smeisler]

Yes, they are different. Hmm, I wonder why fMRIprep would output a gifti that is incompatable....

And yes, the read vs. load surf was a typo on my end.

[smeisler]

I found a compatible .gii from templateflow, and it seems to be past the error now. Thanks!

[damondpham]

Great! If you do want to use fname_gifti_left, another option is resampling it with resample_surf, setting resamp_res=32492.

[smeisler]

When I try that, I get the following error: test <- resample_surf(surf=fname_gifti_left, resamp_res = 32492, hemisphere='left')

Error in read_surf(surf = surf, expected_hemisphere = expected_hemisphere, : The object could not be converted into a surface.
Traceback:

1. resample_surf(surf = fname_gifti_left, resamp_res = 32492, hemisphere = "left")
2. make_surf(gii_post, hemisphere)
3. read_surf(surf = surf, expected_hemisphere = expected_hemisphere, 
 .     resamp_res = NULL)
4. stop("The object could not be converted into a surface.")

And (I cut out most of the error message because it is very big and repetitive, the last part is below): triangle 317355, vertex 0Vertex used twice in one triangle: triangle 317356, vertex 0Vertex used twice in one triangle: triangle 317356, vertex 0Vertex used twice in one triangle: triangle 317357, vertex 0Vertex used twice in one triangle: triangle 317357, vertex 0...

It seems to have that error for every triangle.

I can crosspost this in ciftiTools since this is more of an issue with that.

[damondpham]

Sure, that would be great if you could crosspost to ciftiTools! Would you mind attaching the surface file in question as well? Thanks for helping us iron out these issues.