bug fixes for new v2.3.0

[neurotix93]

Hello

Thank you for the wonderful toolbox!
In the preprocessing step, the tissue analysis is failing with the following error:

2026-03-17 19:58:30 | INFO | pre.preprocess.002 | ├─ FreeSurfer recon-all: ✓ Complete
2026-03-17 19:58:30 | INFO | pre.preprocess.002 | ├─ Tissue analysis: Started
2026-03-17 19:58:30 | ERROR | pre.preprocess.002 | Tissue analysis failed: Labeling.nii.gz not found at /mnt/derivatives/SimNIBS/sub-002/m2m_002/segmentation/Labeling.nii.gz
2026-03-17 19:58:30 | INFO | pre.preprocess.002 | ├─ Tissue analysis: ✗ Failed
2026-03-17 19:58:30 | INFO | pre.preprocess.002 | └─ Pre-processing failed for subject: 002

I wonder if there is a bug, as SimNIBS produces a file called 'labeling.nii.gz' which seems to be the correct file, so maybe there is a case sensitivity issue?

[idossha]

Hi @neurotix93,

Thanks for the kind words and useful feedback!

What version are you using? This is always important to mention as the software been updating very rapidly.

From your screenshot, it looks like it is a path issue.

Version 2.3.0 should be good to go as I tested it both in GUI and scripting.

[neurotix93]

Hello, I just tried it in version 2.3.0 and got the same error.
I think SimNIBS is outputting a file 'labeling.nii.gz' and the toolbox is looking for a case sensitive 'Labeling.nii.gz' -- when I copy the labeling file with a lower case, it seems to work.

[neurotix93]

Im trying on a separate windows install, but I cannot find the .exe file for the windows install anymore on the release. Would it be possible to upload this?

[idossha]

Hi @neurotix93,

Thanks for your input.

1. regarding labeling. Fixed.
2. Regarding executables, it should be available for Linux and Windows within 20 minutes. macOS will be ready later today.

We appreciate any feedback good/bad. v2.3.0 was a massive updating so not surprised there are some edge cases like these. If you continue to encounter anything feel free to continue using this discussion board.

The fix will be retroactive for v2.3.0. No need to reinstall. To apply fix, please close the TI-Toolbox and relaunch. Fixes should be applied once you are logged back in.

Thanks!
Ido

[neurotix93]

Hello Ido,

Thanks so much for replying and applying the fixes so quickly. However, I am now getting an error:

21:20:15| ��OCI runtime exec failed: exec failed: unable to start container process: exec: "simnibs_python": executable file not found in $PATH

[idossha]

I found the issue. Which stems from our recent transition to SimNIBS 4.6. It has been fixed and show be available in 30 minutes.

For the future:

1. If you encounter a loading issue and need to continue working with the TI-Toolbox ASAP, I would recommend using the loader.py script which is the most robust way to load the software.

2. Regarding future issues, printing a one liner is often not enough to debug on our end (though this time was sufficient). It is best to give context. i.e.

• what process was run that resulted in the behavior / what were you trying to do..
• what OS, version, etc...

Thanks again for the feedback and working with me on squashing those little bugs.

Hope you have smooth sailing from here on out.
Ido

p.s. delete the Desktop app you have currently and download the new one. There will be a flag indicating those bugs were fixed in it.

[neurotix93]

Hello,

Thank you so much - it is working now.
However, running into another error whilst preprocessing a single subject's data, just after charm has completed -- using Ubuntu 22.04.5.

Subject atlas segmentation: Started
Running subject_atlas for subject 002 with atlases: a2009s, DK40, HCP_MMP1
Creating a2009s atlas...
usage: subject_atlas [-h] -a {a2009s,DK40,HCP_MMP1} [-o OUT_DIR] [-v] subid
subject_atlas: error: unrecognized arguments: -m
Traceback (most recent call last):
File "", line 198, in _run_module_as_main
File "", line 88, in _run_code
File "/ti-toolbox/tit/pre/main.py", line 43, in
main()
File "/ti-toolbox/tit/pre/main.py", line 23, in main
exit_code = run_pipeline(
^^^^^^^^^^^^^
File "/ti-toolbox/tit/pre/structural.py", line 409, in run_pipeline
_run_subject_pipeline(
File "/ti-toolbox/tit/pre/structural.py", line 99, in _run_subject_pipeline
_run_step(
File "/ti-toolbox/tit/pre/structural.py", line 30, in _run_step
func()
File "/ti-toolbox/tit/pre/structural.py", line 101, in
lambda: run_subject_atlas(
^^^^^^^^^^^^^^^^^^
File "/ti-toolbox/tit/pre/charm.py", line 138, in run_subject_atlas
raise PreprocessError(
tit.pre.utils.PreprocessError: subject_atlas failed for atlas a2009s (exit code 2)
Process returned non-zero exit code (1)

Seems to be an error with the atlas?

[idossha]

Strange. Looking into it. What machine are you running it on?

I think it is defnitely the new SimNIBS version. This is something I am still figuring out myself actually. They changed their numpy dependency to a significantly newer version, but there is some strange behavior that I thought I resolved.

I am looking into it.

[idossha]

This is a SimNIBS 4.6.0 packaging bug, not a TI-Toolbox bug. SimNIBS 4.6's source code uses np.nan_to_num(..., copy=None), a numpy 2.x feature - but their installer seems to ship numpy 1.26.4, which only accepts True or False for the copy parameter. The crash happens at the very end of CHARM when it calls nifti_transform() to write the label image from the finished mesh.

I wrote a patch for this and it should actually be solving this. Are you running the code locally or within the container environment of the TI-Toolbox?

Last I test the full preprocessing pipeline all worked well. I am testing it again and should have some more input soon.

[neurotix93]

I was running this locally in the container; on a linux machine.

[idossha]

So I was not able to reproduce it. It worked well on my end:

See end of log:

[ simnibs ] INFO: Reading settings from /mnt/000/derivatives/SimNIBS/sub-ernie/m2m_ernie/settings.ini
[ simnibs ] INFO: Reading settings from /root/SimNIBS-4.6/simnibs_env/lib/python3.11/site-packages/simnibs/charm.ini
[ simnibs ] INFO: charm run finished: Mon Mar 23 22:18:27 2026
[ simnibs ] INFO: Total running time: 01:11:32
SimNIBS charm: ✓ Complete
Subject atlas segmentation: Started
Running subject_atlas for subject ernie with atlases: a2009s, DK40, HCP_MMP1
Creating a2009s atlas...
[ simnibs ] INFO: Writing: /mnt/000/derivatives/SimNIBS/sub-ernie/m2m_ernie/segmentation/lh.ernie_a2009s.annot
[ simnibs ] INFO: Writing: /mnt/000/derivatives/SimNIBS/sub-ernie/m2m_ernie/segmentation/rh.ernie_a2009s.annot
Creating DK40 atlas...
[ simnibs ] INFO: Writing: /mnt/000/derivatives/SimNIBS/sub-ernie/m2m_ernie/segmentation/lh.ernie_DK40.annot
[ simnibs ] INFO: Writing: /mnt/000/derivatives/SimNIBS/sub-ernie/m2m_ernie/segmentation/rh.ernie_DK40.annot
Creating HCP_MMP1 atlas...
[ simnibs ] INFO: Writing: /mnt/000/derivatives/SimNIBS/sub-ernie/m2m_ernie/segmentation/lh.ernie_HCP_MMP1.annot
[ simnibs ] INFO: Writing: /mnt/000/derivatives/SimNIBS/sub-ernie/m2m_ernie/segmentation/rh.ernie_HCP_MMP1.annot
All atlases created successfully for subject ernie
Subject atlas segmentation: ✓ Complete
Pre-processing completed successfully for subject: ernie

Can you try and exit the ti-toolbox and give charm another try?

[neurotix93]

Hello Ido,

Thanks so much! I created a new project folder and it worked. No error message :)

All the best!

[neurotix93]

Here is potentially another small bug with QSIPrep -- as I want to incorporate anisotropic conductivities into the modelling.
Have only selected QSIPrep, QSIRecon and Extract DTI tensor for SimNibs. All on a linux machine, desktop app.

Beginning pre-processing for subject: 002
QSIPrep DWI preprocessing: Started
Starting QSIPrep for subject 002
Pulling Docker image pennlinc/qsiprep:1.1.1...
Failed to pull pennlinc/qsiprep:1.1.1: failed to connect to the docker API at unix:///var/run/docker.sock; check if the path is correct and if the daemon is running: dial unix /var/run/docker.sock: connect: no such file or directory
Traceback (most recent call last):
File "", line 198, in _run_module_as_main
File "", line 88, in _run_code
File "/ti-toolbox/tit/pre/main.py", line 43, in
main()
File "/ti-toolbox/tit/pre/main.py", line 23, in main
exit_code = run_pipeline(
^^^^^^^^^^^^^
File "/ti-toolbox/tit/pre/structural.py", line 409, in run_pipeline
_run_subject_pipeline(
File "/ti-toolbox/tit/pre/structural.py", line 138, in _run_subject_pipeline
_run_step(
File "/ti-toolbox/tit/pre/structural.py", line 30, in _run_step
func()
File "/ti-toolbox/tit/pre/structural.py", line 140, in
lambda: run_qsiprep(
^^^^^^^^^^^^
File "/ti-toolbox/tit/pre/qsi/qsiprep.py", line 128, in run_qsiprep
raise PreprocessError(
tit.pre.utils.PreprocessError: Failed to pull QSIPrep image: pennlinc/qsiprep:1.1.1
Process returned non-zero exit code (1)

Also, I cannot see if the TIToolbox call to QSIPrep allows the use of reverse phase encoded data stored in a directory like fmap for distortion correction? Is that possible?

[idossha]

can you move this to a new thread please? this is a pretty big topic so worth its own discussion.

[idossha]

In the meantime, here are the answers to your questions:

1. Good catch. the qsiprep/recon was not fully implemented in the compiled linux version which is now fixed. Please delete your latest desktop and reinstall. This is one of those times that you have to reinstall.

2. Reverse Phase-Encoded Data / Fieldmaps

Yes, this is supported. TI-Toolbox mounts your entire project directory into a QSIPrep container under the hood, so if your data follows the BIDS specification with an fmap/ directory, QSIPrep will automatically discover and use those fieldmaps for susceptibility distortion correction - no extra configuration needed on the TI-Toolbox side.

Your BIDS structure should look something like:

project_root/
├── sub-002/
│ ├── dwi/
│ │ ├── sub-002_dwi.nii.gz
│ │ ├── sub-002_dwi.bval
│ │ └── sub-002_dwi.bvec
│ └── fmap/
│ ├── sub-002_dir-AP_epi.nii.gz
│ ├── sub-002_dir-AP_epi.json
│ ├── sub-002_dir-PA_epi.nii.gz
│ └── sub-002_dir-PA_epi.json

Additionally, TI-Toolbox exposes QSIPrep's --distortion-group-merge parameter (options: concatenate, average, or none) for controlling how distortion groups are combined, which can be set in the preprocessing configuration.

In essense, if you are familier with qsiprep and qsirecon workflows, it is exactly the same. However, as this is a very new feature to TI-Toolbox, if you encouter any weird behvaior, especially if you have experience with those pipelines, let us know.

For more details, see:

• QSIPrep fieldmap handling: https://qsiprep.readthedocs.io/en/latest/preprocessing.html
• TI-Toolbox diffusion processing docs: https://idossha.github.io/TI-Toolbox/api/pipeline/preprocessing/

[neurotix93]

That's great, thanks so much. Am trying it now. If any issues crop up, will start a new thread and let you know.

[idossha]

Awesome. Glad it all worked out. Again, if anything comes up, always feel free to reach out with an update or just for QA.

If you find we are missing some features you need internally we are happy to try and look into things to make it a more complete toolbox! If you find this project helpful please star our repo to help us grow!

Best,
Ido

[neurotix93]

Thank you so much :) I will!

[idossha]

Hi @neurotix93 — quick update: several preprocessing/QSI follow-ups related to this thread have now been addressed and are queued for v2.3.1.

The release will include preprocessing cleanup, better GUI process/error reporting, and an upgrade of QSIPrep/QSIRecon to PennLINC 26.0.0 with compatibility handling for the current CLI (concat for QSIPrep and --input-type qsiprep for QSIRecon). The Docker socket mount for spawning QSI sibling containers is also part of the maintained compose setup.

These updates should be released with v2.3.1 in the coming days. If anything persists after updating, please share the new log output and I’ll follow up from there.

p.s. Would love to hear if you ran any comparison with previous DTI pipelines and can share either qualitative or quantitative results.

Best,
Ido

[neurotix93]

Dear Ido,

That's great. Thank you so much! I will test it out, and try some comparisons too.

Laith