Closed eds-slim closed 11 months ago
@effigies, could we exclude N3 from the freesurfer workflow, since we are running N4 ourselves?
You'll need to check whether nu.mgz
is aligned to orig.mgz
. It takes talairach.xfm
as an input: https://surfer.nmr.mgh.harvard.edu/fswiki/ReconAllTableStableV6.0
Thanks for taking this up! Is there anything I can test further or otherwise contribute from my side?
@oesteban @effigies: I looked into the issue a bit further and tried to run mri_convert
on the offending orig.mgz
outside the container, this works without error.
As far as I can tell, the local and containerized versions of mri_convert
are the same
mri_convert --version
mri_convert.bin --version
stable6
-bash-4.2$ singularity shell --userns $WORK/containers/fmriprep/sandbox
Singularity sandbox:~> mri_convert --version
mri_convert.bin --version
stable6
I do note that the bc
version in the container is 1.06.95
, which is older than my local verison 1.07.1
. Could this be contributing to the problem?
Thanks for further input to resolve this.
The same error
mri_convert orig.mgz ./tmp.mri_nu_correct.mni.6097/nu0.mnc -odt float
mri_convert.bin orig.mgz ./tmp.mri_nu_correct.mni.6097/nu0.mnc -odt float
/opt/freesurfer/bin/mri_convert: line 3: 6272 Segmentation fault mri_convert.bin "$@"
ERROR: converting to minc
occurs with the latest smriprep
version 2019-09-09-76bc290c67e9
, i.e. smriprep v0+unknown
.
After further scrutiny I have been able to use the -nocache
option to volume_stats
as described here to get the freesurfer recon-all
command to execute a little bit further.
It now crashes out with a segmentation fault on spline_smooth
. I will raise this issue with the N3 people and report back when I get a response. The alternative suggested by @oesteban, i.e. to skip N3 altogether, would be even more appealing. How could that be achieved?
Sorry, I'm still just not understanding why this is happening. I don't see any reason that N3 should fail after N4 runs.
Is the issue that we're inducing negative numbers in the image with N4 correction? What do these images look like?
If you're able to share the original T1w image, I could see if I can reproduce this.
@eds-slim Hello, did you find a solution? I got exactly the same issue right now.
I face a similar issue. I could run a few weeks back recon-all (freesurfer6.0) from Singularity on our server. Trying the same container with the same subject, it exits with the error above. Checking the fs log files from the previous successful run and the current one, no difference (until the line where the error is mentioned). However, when I check the log files in the individual folder/mri/mri_nu_correct.mni.log) there are some differences: from the successful run: current, bad run: It seems that the mri_nu_correct.mni called with different arguments.
It seems that we have had some update on our server and there was some issue with mounting the tmp folder in the container. It seems that freesurfer could not write there, so in the next step it could not find data. Sorry for the noise.
No worries, glad you figured it out. Closing this as there's nothing actionable. A new issue can be opened if it resurfaces.
Hi, when using fmriprep version v1.5.0rc1 in singularity version 3.4.2-rc-1 I receive an error in recon-all
Specifically, there appears to be an segmentation fault in
mri_convert
I get the same error, when I run this command isolated through singularity:
Interestingly, no error occurs when i change
-odt float
to-odt uchar
or-odt short
. The seg fault does occur for-odt int
.Any ideas? Thanks.