Hey @sbajaj1 and thanks for checking out LST-AI!

So Step 6.1 downloads and installs a pre-compiled version of greedy (that we compiled on Ubuntu), so I am unsure whether that pre-compiled version would work on Mac OS. I would recommend to just give it a try, i.e. use a T1w and FLAIR and run lst. Please report if that works for you 👍

If you encounter any problems, we might use Step 6.2 (compiling greedy on MAC) instead of Step 6.1. However, I have not tested on MAC (and do not have a MAC), so we might need to troubleshoot that together if you run into any problems.

Hi Julian,

Thanks for your quick response. I am having serious trouble in installing this toolbox (may be because the instructions are not valid for Mac). Please see my steps below after starting from scratch. I would greatly appreciate your help resolving this:

1. apt-get does not work for Mac, so I ran:
   brew update && brew install git wget unzip python3
   (all of the above are now installed, including pip3 which is I think part of python3)

2. Following set of commands work fine:
   mkdir lst_directory
   cd lst_directory
   python3 -m venv /Users/sbajaj/
   source /Users/sbajaj/bin/activate

3. Following set of commands work fine:
   git clone https://github.com/CompImg/LST-AI/
   cd LST-AI

but pip install -e . gives me following error, which I am unable to resolve:
ERROR: Ignored the following versions that require a different python version: 1.21.2 Requires-Python >=3.7,<3.11; 1.21.3 Requires-Python >=3.7,<3.11; 1.21.4 Requires-Python >=3.7,<3.11; 1.21.5 Requires-Python >=3.7,<3.11; 1.21.6 Requires-Python >=3.7,<3.11; 1.6.2 Requires-Python >=3.7,<3.10; 1.6.3 Requires-Python >=3.7,<3.10; 1.7.0 Requires-Python >=3.7,<3.10; 1.7.1 Requires-Python >=3.7,<3.10; 1.7.2 Requires-Python >=3.7,<3.11; 1.7.3 Requires-Python >=3.7,<3.11; 1.8.0 Requires-Python >=3.8,<3.11; 1.8.0rc1 Requires-Python >=3.8,<3.11; 1.8.0rc2 Requires-Python >=3.8,<3.11; 1.8.0rc3 Requires-Python >=3.8,<3.11; 1.8.0rc4 Requires-Python >=3.8,<3.11; 1.8.1 Requires-Python >=3.8,<3.11
ERROR: Could not find a version that satisfies the requirement tensorflow<2.12.0 (from lst-ai) (from versions: 2.13.0rc0, 2.13.0rc1, 2.13.0rc2, 2.13.0, 2.13.1, 2.14.0rc0, 2.14.0rc1, 2.14.0, 2.14.1, 2.15.0rc0, 2.15.0rc1, 2.15.0)
ERROR: No matching distribution found for tensorflow<2.12.0

I then updated my python to 3.9, still I get similar error about tensor flow as following:

ERROR: Could not find a version that satisfies the requirement tensorflow<2.12.0 (from lst-ai) (from versions: 2.13.0rc0, 2.13.0rc1, 2.13.0rc2, 2.13.0, 2.13.1, 2.14.0rc0, 2.14.0rc1, 2.14.0, 2.14.1, 2.15.0rc0, 2.15.0rc1, 2.15.0)
ERROR: No matching distribution found for tensorflow<2.12.0

Thank you for the debugs. Can you try relaxing the version restrictions for tensorflow in setup.py?

From:

      install_requires = [
        'numpy<1.24.0',
        'pillow<10.1.0',
        'scipy>=1.9.0',
        'scikit-image>=0.21.0',
        'tensorflow<2.12.0',
        'tensorflow-addons<0.20.0',
        'nibabel>=4.0.0',
        'requests>=2.20.0'
      ],

To e.g.

      install_requires = [
        'numpy<1.24.0',
        'pillow<10.1.0',
        'scipy>=1.9.0',
        'scikit-image>=0.21.0',
        'tensorflow',
        'tensorflow-addons',
        'nibabel>=4.0.0',
        'requests>=2.20.0'
      ],

Hi Julian,

That has partially resolved the issue- thank you so much for that. After I edited setup.py file as you suggested I get the following warning after I check if lst is installed fine:

/Users/sbajaj/lib/python3.9/site-packages/tensorflow_addons/utils/tfa_eol_msg.py:23: UserWarning:

TensorFlow Addons (TFA) has ended development and introduction of new features.
TFA has entered a minimal maintenance and release mode until a planned end of life in May 2024.
Please modify downstream libraries to take dependencies from other repositories in our TensorFlow community (e.g. Keras, Keras-CV, and Keras-NLP).

For more information see: tensorflow/addons#2807

warnings.warn(
###########################

Thank you for using LST-AI. If you publish your results, please cite our paper:
Wiltgen T, McGinnis J, Schlaeger S, Voon C, Berthele A, Bischl D, Grundl L, Will N, Metz M, Schinz D, Sepp D,Prucker P, Schmitz-Koep B, Zimmer C, Menze B, Rueckert D, Hemmer B, Kirschke J, Muehlau M, Wiestler B. LST-AI: a Deep Learning Ensemble for Accurate MS Lesion Segmentation. medRxiv preprint 10.1101/2023.11.23.23298966, 2023.
###########################

usage: lst [-h] --t1 T1 --flair FLAIR --output OUTPUT [--existing_seg EXISTING_SEG] [--temp TEMP] [--segment_only] [--annotate_only] [--stripped] [--threshold THRESHOLD] [--fast-mode] [--device DEVICE]
[--threads THREADS]

If I ignore the above warning and run lst command as following, I get error:

$ lst --t1 /Users/sbajaj/BRAINIX_NIFTI_T1.nii.gz --flair /Users/sbajaj/BRAINIX_NIFTI_FLAIR.nii.gz --output /Users/sbajaj/check_output

/Users/sbajaj/lib/python3.9/site-packages/tensorflow_addons/utils/tfa_eol_msg.py:23: UserWarning:

TensorFlow Addons (TFA) has ended development and introduction of new features.
TFA has entered a minimal maintenance and release mode until a planned end of life in May 2024.
Please modify downstream libraries to take dependencies from other repositories in our TensorFlow community (e.g. Keras, Keras-CV, and Keras-NLP).

For more information see: tensorflow/addons#2807

warnings.warn(
###########################

Thank you for using LST-AI. If you publish your results, please cite our paper:
Wiltgen T, McGinnis J, Schlaeger S, Voon C, Berthele A, Bischl D, Grundl L, Will N, Metz M, Schinz D, Sepp D,Prucker P, Schmitz-Koep B, Zimmer C, Menze B, Rueckert D, Hemmer B, Kirschke J, Muehlau M, Wiestler B. LST-AI: a Deep Learning Ensemble for Accurate MS Lesion Segmentation. medRxiv preprint 10.1101/2023.11.23.23298966, 2023.
###########################

Looking for model weights in /Users/sbajaj/bin.
Images need to be skull-stripped. Processing with HD-BET.
Traceback (most recent call last):
File "/Users/sbajaj/bin/lst", line 269, in
mni_registration(t1w_atlas,
File "/Users/sbajaj/lst_directory/LST-AI/LST_AI/register.py", line 40, in mni_registration
subprocess.run(shlex.split(rigid_call), check=True)
File "/opt/homebrew/Cellar/[email protected]/3.9.18_1/Frameworks/Python.framework/Versions/3.9/lib/python3.9/subprocess.py", line 505, in run
with Popen(*popenargs, **kwargs) as process:
File "/opt/homebrew/Cellar/[email protected]/3.9.18_1/Frameworks/Python.framework/Versions/3.9/lib/python3.9/subprocess.py", line 951, in init
self._execute_child(args, executable, preexec_fn, close_fds,
File "/opt/homebrew/Cellar/[email protected]/3.9.18_1/Frameworks/Python.framework/Versions/3.9/lib/python3.9/subprocess.py", line 1837, in _execute_child
raise child_exception_type(errno_num, err_msg, err_filename)
OSError: [Errno 8] Exec format error: 'greedy'

I am not getting error while installing greedy - but still even after having lst command showing OK in my terminal, the analysis command is showing above error.

Good morning @sbajaj1, thanks for doing all of the debugging. That's super good for troubleshooting. And thank you for being so patient with the installation process, we did not for see MAC Users installing LST-AI natively that soon, and developed LST-AI for the Linux platforms.

In general, and perhaps I should have mentioned this earlier as in retrospect perhaps that was not ultimately clear. (If so, please let me know and I will add this to the README.md. Also, sincere apologies!). You should always be able to use the docker containers (provided in CPU and GPU) to build and run dockerized versions of LST-AI instead of natively installing it on your system. For MAC OSX, I would use the CPU-version as Nvidia GPUs are likely not supported.

If you want to continue with the native installation of LST-AI on MAC OS, I have the following advice:

Likely, the installed version of greedy (which we compiled on Ubuntu) does not work on MAC OS. To test that, we can try the following. Could you please open a new terminal and call greedy ? It should display the options menu, e.g.:

(base) juli@juli-Latitude-7400:~$ greedy
greedy: Paul's greedy diffeomorphic registration implementation
Usage: 
  greedy [options]
Required options: 
  -d DIM                 : Number of image dimensions
  -i fix.nii mov.nii     : Image pair (may be repeated)
  -o <file>              : Output file (matrix in affine mode; image in deformable mode,     
[...]
Environment variables: 
  GREEDY_DATA_ROOT       : if set, filenames can be specified relative to this path

If that is not the case, then you will need to download a different version of greedy - which fortunately is available for MAC OS, and you can install / download the package for MAC OS. You would then replace the binary you placed in step 6.1. (II) with the binary you just downloaded.

Did that help?

Cheers,
Julian

**Hi Julian,

Thanks for the details about next steps. When I try docker, I am getting the following error:

[ 7/25] RUN pip install -e .:
0.257 Obtaining file:///custom_apps/lst_directory/LST-AI
0.258 Preparing metadata (setup.py): started
0.319 Preparing metadata (setup.py): finished with status 'done'
0.453 Collecting nibabel>=4.0.0
0.589 Downloading nibabel-5.1.0-py3-none-any.whl (3.3 MB)
1.833 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 3.3/3.3 MB 2.7 MB/s eta 0:00:00
2.149 Collecting numpy<1.24.0
2.173 Downloading numpy-1.23.5-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl (13.9 MB)
4.780 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 13.9/13.9 MB 6.0 MB/s eta 0:00:00
5.134 Collecting pillow<10.1.0
5.159 Downloading Pillow-10.0.1-cp310-cp310-manylinux_2_28_aarch64.whl (3.5 MB)
5.616 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 3.5/3.5 MB 7.6 MB/s eta 0:00:00
5.684 Collecting requests>=2.20.0
5.705 Downloading requests-2.31.0-py3-none-any.whl (62 kB)
5.712 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 62.6/62.6 KB 11.2 MB/s eta 0:00:00
5.821 Collecting scikit-image>=0.21.0
5.845 Downloading scikit_image-0.22.0-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl (14.1 MB)
7.358 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 14.1/14.1 MB 9.5 MB/s eta 0:00:00
7.569 Collecting scipy>=1.9.0
7.863 Downloading scipy-1.11.4-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl (32.9 MB)
10.28 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 32.9/32.9 MB 12.6 MB/s eta 0:00:00
10.37 ERROR: Could not find a version that satisfies the requirement tensorflow-addons<0.20.0 (from lst-ai) (from versions: none)
10.37 ERROR: No matching distribution found for tensorflow-addons<0.20.0

Dockerfile:20

18 | RUN git clone https://github.com/CompImg/LST-AI/
19 | WORKDIR /custom_apps/lst_directory/LST-AI
20 | >>> RUN pip install -e .
21 |
22 | # Setup for greedy (Choose either pre-compiled or compilation from source)

ERROR: failed to solve: process "/bin/sh -c pip install -e ." did not complete successfully: exit code: 1**

@sbajaj1 Oh no, we are running into the "chicken or the egg" problem - i.e., if you use MAC OS to build the docker container, then the installation on MAC OS needs to work, which was why we turned to docker in the first place.

However, we can bypass this problem if I share the pre-built docker container with you. So I just built the newest version of the docker (CPU) container on our Ubuntu server, and I am currently uploading it to dockerhub (which unfortunately takes some time as it's 12GB large).

The next stops will be:

1. Download (i.e. pull the docker container from our official dockerhub repo)
2. Run the docker container with your files.

However, I will give you the exact instructions to do so once the upload has completed.

I am deeply sorry that we did not provide the docker container in the first place; I did not consider that you need a working solution for MAC to build the docker container. Thank you very much for your patience, and I think that we will have finished the installation process for you, and you will have a working LST-AI solution for your MAC laptop.

Thank you for your understanding,

I will give you some more instructions soon once the upload has completed! 🙂

Have a nice evening,
Julian

Great - thank you so much Julian.

I will wait for the newest version and the next steps. Thank you so much for all your help !!

Thanks for your patience @sbajaj1 - the next steps should hopefully fix the issues and allow you to run LST-AI on your MAC.

First things first, I compiled a CPU docker container, so it won't be using your MAC's GPU, as we don't support that (yet). Also, CPU docker's are much easier to handle as we cannot really forsee which GPU hardware our user base is going to leverage (we would even need different GPU docker versions for Nvidia GPU / CUDA Versions).

Consequently, segmentation will take significantly longer (in the range of minutes on a CPU rather than seconds on the GPU). Including the registration and skull-stripping I expect it to be >10mins.

However, the advantage should be that it actually works 🙂 - and I will work on a GPU-MAC version if I can get one in my hands any time soon (e.g. from a colleague).

That being said, and without further ado, here is the workflow for installing the CPU-docker on your system:

1. Please open a terminal and run the following command:

docker build -t jqmcginnis/lst-ai_cpu:latest .

This retrieves the pre-built docker container from dockerhub.

2. Run the updated docker instructions from here (not the ones from the Readme):

docker run -v /home/ginnis/lst_in:/custom_apps/lst_input -v /home/ginnis/lst_out/:/custom_apps/lst_output -v /home/ginnis/lst_temp/:/custom_apps/lst_temp jqmcginnis/lst-ai_cpu:latest --t1 /custom_apps/lst_input/t1.nii.gz --flair /custom_apps/lst_input/flair3d.nii.gz --output /custom_apps/lst_output --temp /custom_apps/lst_temp --device cpu

Some more notes:

• You might need to add yourself to the docker group before you run the code (otherwise, you need to run everything with sudo, which you can do, but I would not recommend as a general practise).
• Importantly, you should not forget the --device cpu instruction, as this will tell LST-AI to run your code on the CPU. 0 or any other GPU will throw an error that you don't have the Nvidia drivers installed.
• The image in-/out-/temp paths e.g. /home/ginnis/lst_in should already exist, you can replace them with anything you like (and modify the input images' names accordingly).

Hope this helped, and please let me know, we will update the Readme.md according to your experiences once we have a working solution for MAC.

Thanks again,
Julian

Hi Julian,

Ok so I ran the two commands you wrote. The first one, it was showing some error, so I installed docker app on my Mac Pro and ran jqmcginnis/lst-ai_cpu:latest. It seems docker is running lst-ai_cpu fine. Please see attached screenshot.

Then I ran the following command in terminal:

docker run -v lst_in:/custom_apps/lst_input -v lst_out:/custom_apps/lst_output -v lst_temp:/custom_apps/lst_temp jqmcginnis/lst-ai_cpu:latest --t1 /Users/sbajaj/sample_test/BRAINIX_NIFTI_T1.nii.gz --flair /Users/sbajaj/sample_test/BRAINIX_NIFTI_FLAIR.nii.gz --output /Users/sbajaj/lst_output --temp /Users/sbajaj/lst_temp --device cpu

But this gives me following warning:
WARNING: The requested image's platform (linux/amd64) does not match the detected host platform (linux/arm64/v8) and no specific platform was requested

(however, adding --platform linux/amd64 to the command got rid of this above warning)

and then shows
The TensorFlow library was compiled to use AVX instructions, but these aren't available on your machine.

And now, its been forever, I do not see any activity on my terminal, please see attached screenshot.

Docker status screenshot is also attached.

I am not sure if I am running my docker run command correctly or not?

Hi @sbajaj1

Thanks for staying with us here, and I am sorry that the process of getting this up and running on MAC OS is so utterly painful. I expected it to be a lot easier with docker, as tensorflow and pytorch (and other python libraries) are cross-platform (and they work out of the box in many projects), however in our specific case it seems to be particularly hard.

I dug deeper, and it feels like the issue is centered around tensorflow / tensorflow-addons incompatibility issues with MAC-OS. This GitHub issue includes a rather lengthy thread regarding the problems we are also running into. I tried building the docker container on my ubuntu laptop with the --platform linux/amd64,linux/arm64/v8 flag, but the build process fails even if I relax the package constraints as we did earlier:

#0 56.42 Collecting tensorflow
#0 56.46   Downloading tensorflow-2.15.0-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl (2.2 kB)
#0 57.02 ERROR: Could not find a version that satisfies the requirement tensorflow-addons (from lst-ai) (from versions: none)
#0 57.02 ERROR: No matching distribution found for tensorflow-addons
------
Dockerfile:20
--------------------
  18 |     RUN git clone https://github.com/jqmcginnis/LST-AI/
  19 |     WORKDIR /custom_apps/lst_directory/LST-AI
  20 | >>> RUN pip install -e .
  21 |     
  22 |     # Setup for greedy (Choose either pre-compiled or compilation from source)
--------------------
error: failed to solve: process "/dev/.buildkit_qemu_emulator /bin/sh -c pip install -e ." did not complete successfully: exit code: 1

Moreover, I have found that tensorflow-addons in particular is difficult to get up and running on MAC-OS, as documented here. While this makes me hopeful that it is possible, I would likely need access to a MAC-OS laptop to test and getting everything up and running myself.

With this in mind, I reached out and asked a collaborator (who is working on MAC OS, and is very experienced with docker) for help and we will meet on Friday. Thus, I would propose that I(we do some more debugging on our end, and report back to you, once I know more.

In the meantime, please let me know if I can assist with the segmentation. E.g. if you have a tight schedule and need segmentations now, I would be willing to run lst-ai for you, if you are able to share data (naturally over a private channel and not via github).

Hi Julian,

That really explains the issues.

Thank you so much once again for your willingness and hard work in getting this resolved. Also, thanks for offering help with segmentation at your end. Good news is that I can definitely wait until next week (or more if needed) as I do not have specific deadline. I will just put this project on brief hold without any issue. So, please take your time!

I am looking forward to receiving the next update and further using this LST-AI method.

Best,
Sahil

Hi @sbajaj1

Thank you very much for your patience. With the help of Florian Kofler (Helmholtz), I have been able to narrow it down to the unavailability of Tensorflow-Addons for the arm64 platform, which, frankly speaking, is very surprising to me. Apparently, Tensorflow-Addons has never been supported for both linux and MAC-OS ARM architectures.

However, we are not giving up, and I will check in the upcoming days if we can replace the functionality we have been using from this library with (1) and own implementation or (2) if I can compile tensorflow-addons for arm64 platforms by ourselves.

Just wanted to give you a quick update and let you know that we are (still) working (with highest priority) on a solution to make LST-AI work on your, and potentially other, ARM64 platforms.

Have a nice day,

Julian

yes, we could reproduce your issue and are working on a fix.

Currently, there seem to be three options:

1. build tfa for arm linux
2. remove tfa from the tech stack
3. build the docker image with a different base image

I am curious about which solution will win in the end :)

Hi @sbajaj1

Thank you very much for your patience. With the help of Florian Kofler (Helmholtz), I have been able to narrow it down to the unavailability of Tensorflow-Addons for the arm64 platform, which, frankly speaking, is very surprising to me. Apparently, Tensorflow-Addons has never been supported for both linux and MAC-OS ARM architectures.

However, we are not giving up, and I will check in the upcoming days if we can replace the functionality we have been using from this library with (1) and own implementation or (2) if I can compile tensorflow-addons for arm64 platforms by ourselves.

Just wanted to give you a quick update and let you know that we are (still) working (with highest priority) on a solution to make LST-AI work on your, and potentially other, ARM64 platforms.

Have a nice day,

Julian

Great - thank you so much Julian @jqmcginnis and Florian @neuronflow for the update and working on this. That sounds like a plan. I am looking forward to getting the next updates.

@sbajaj1 It seems that just right in time for Christmas, we might have a little 🎁

For the last two weeks, @neuronflow and I have been debugging multiple issues arising with MacOS / ARM64 Dockerfiles, and I am happy to share that we have made quite some progress:

1. We have substituted Tensorflow-Addons (tfa) by a custom implementation, no longer requiring it in the installation process. We will merge it to this repo after some final tests, the code is currently hosted in a fork.
2. Now, we build greedy natively in our dockerfile which took us quite some time to make it work on ARM64 as well.
3. We switched to a slim base image, which might be more efficient as it has less overhead.
4. We tested everything on a M1 Max, let's hope it works as seamlessly on an M2 Max as well 🍀

To obtain the new LST-AI image, please do the following:

docker pull jqmcginnis/lst-ai_cpu:latest

On your MAC OS platform, this should automatically fetch the ARM64 version (linux/amd64 would be available as well, but not suited for your platform).

Then, you can finally run:

docker run -v /Users/nf/projects/lst/input:/custom_apps/lst_input -v /Users/nf/projects/lst/output:/custom_apps/lst_output -v /Users/nf/projects/lst/temp:/custom_apps/lst_temp jqmcginnis/lst-ai_cpu:latest  --t1 /custom_apps/lst_input/T1W.nii.gz --flair /custom_apps/lst_input/FLAIR.nii.gz --output /custom_apps/lst_output --temp /custom_apps/lst_temp --device cpu 

Please let us know if this works for you.

We will update the Dockerfile and Release soon, once we have obtained your feedback.
And finally, a big hank you to @neuronflow again, and thank you for your patience @sbajaj1!

Cheers,
Julian

Hi Julian,

Thank you so much for working on this. Sorry for the late response (I was on Christmas break). The docker pull command ran fine without any issue and the download was successful, but the docker run command is giving the following error. It seems there is some minor issue (may be from my side):

sudo docker run -v /Users/nf/projects/lst/input:/custom_apps/lst_input -v /Users/nf/projects/lst/output:/custom_apps/lst_output -v /Users/nf/projects/lst/temp:/custom_apps/lst_temp jqmcginnis/lst-ai_cpu:latest --t1 /Users/sbajaj/sample_test/BRAINIX_NIFTI_T1.nii.gz --flair /Users/sbajaj/sample_test/BRAINIX_NIFTI_FLAIR.nii.gz --output /custom_apps/lst_output --temp /custom_apps/lst_temp --device cpu
Password:
docker: Error response from daemon: error while creating mount source path '/host_mnt/Users/nf/projects/lst/input': mkdir /host_mnt/Users/nf: permission denied.
ERRO[0000] error waiting for container:

does the same happen when you ran without sudo?

/host_mnt/Users/nf/projects/lst/input

Does this folder and the others you mount exist already? What are their permissions?

Hi @neuronflow

Yes, I get the same error without sudo as well. No, these folders do not exist already. And, permissions are fine too as I am the only admin. May be something wrong in the command. Just to reiterate, I am doing the following (my data T1w and FALIR images are at the location: /Users/sbajaj/sample_test/....nii.gz

MDACM0CL9004107:~ sbajaj$ docker pull jqmcginnis/lst-ai_cpu:latest

latest: Pulling from jqmcginnis/lst-ai_cpu
Digest: sha256:76d1108fae4835f32f41852f5c27492b94ef39d9696eec06cfed5c0086857a2d
Status: Image is up to date for jqmcginnis/lst-ai_cpu:latest
docker.io/jqmcginnis/lst-ai_cpu:latest

MDACM0CL9004107:~ sbajaj$ docker run -v /Users/nf/projects/lst/input:/custom_apps/lst_input -v /Users/nf/projects/lst/output:/custom_apps/lst_output -v /Users/nf/projects/lst/temp:/custom_apps/lst_temp jqmcginnis/lst-ai_cpu:latest --t1 /Users/sbajaj/sample_test/BRAINIX_NIFTI_T1.nii.gz --flair /Users/sbajaj/sample_test/BRAINIX_NIFTI_FLAIR.nii.gz --output /Users/sbajaj/lst_output --temp /Users/sbajaj/lst_temp --device cpu

docker: Error response from daemon: error while creating mount source path '/host_mnt/Users/nf/projects/lst/input': mkdir /host_mnt/Users/nf: permission denied.
ERRO[0000] error waiting for container:

Hey @sbajaj1 - Happy New Year ✨

Thank you for the helpful debugging - I think there are some misunderstandings with the volume mounting process, i.e. wrong mounting of paths (/Users/nf/projects/). This here is the initial command:

docker run -v /Users/nf/projects/lst/input:/custom_apps/lst_input -v /Users/nf/projects/lst/output:/custom_apps/lst_output -v /Users/nf/projects/lst/temp:/custom_apps/lst_temp jqmcginnis/lst-ai_cpu:latest  --t1 /custom_apps/lst_input/T1W.nii.gz --flair /custom_apps/lst_input/FLAIR.nii.gz --output /custom_apps/lst_output --temp /custom_apps/lst_temp --device cpu

You should replace the /Users/nf/projects/lst/input, /Users/nf/projects/lst/output, /Users/nf/projects/lst/temp with paths on your system that already exist, e.g. Users/sbajaj/sample_test. The latter parts of the code should remain unchanged, as the docker container looks for these files in the specific directories, e.g. /custom_apps/lst_input/, /custom_apps/lst_output/ and /custom_apps/lst_temp/. Only the filename (e.g. BRAINIX_NIFTI_FLAIR.nii.gz) must be inserted. Can you please try this here:

docker run -v /Users/sbajaj/sample_test/:/custom_apps/lst_input -v /Users/sbajaj/lst_output:/custom_apps/lst_output -v /Users/sbajaj/lst_temp:/custom_apps/lst_temp jqmcginnis/lst-ai_cpu:latest  --t1 /custom_apps/lst_input/BRAINIX_NIFTI_T1.nii.gz --flair /custom_apps/lst_input/BRAINIX_NIFTI_FLAIR.nii.gz --output /custom_apps/lst_output --temp /custom_apps/lst_temp --device cpu 

Let me know if this works 👍

Cheers,
Julian

Happy new year too! 🚀
Yes, make sure that the folders you mount already exist on your host system. Please report back with what you get when running Julian's command :)

Hi @neuronflow @jqmcginnis

Now the command is working fine without any error. Thank you so much !

Just a follow-up quick question: Several files are generated (see attached)

and are now saved in lst_temp folder (though nothing is saved in lst_output). Is there any manual/description available about the details of the generated files?
Also the LST-SPM toolbox generates lesion mask and volume/size of the lesion. I was wondering if LST_AI also somehow saves those lesion mask, volume/size files?

Thanks,
Sahil

Can you copy-paste your console output, please?

The computation should take a while. The final results should appear in the output folder.

The temp folder should contain intermediate steps generated on the way toward populating the output folder.

PS: lesion size should be trivial to implement by counting voxels?

Hi @neuronflow

Here you go (I just noticed this error in this output: vnl_lbfgs: Error. Netlib routine lbfgs failed.):

MDACM0CL9004107:~ sbajaj$ docker run -v /Users/sbajaj/sample_test/:/custom_apps/lst_input -v /Users/sbajaj/lst_output:/custom_apps/lst_output -v /Users/sbajaj/lst_temp:/custom_apps/lst_temp jqmcginnis/lst-ai_cpu:latest --t1 /custom_apps/lst_input/BRAINIX_NIFTI_T1.nii.gz --flair /custom_apps/lst_input/BRAINIX_NIFTI_FLAIR.nii.gz --output /custom_apps/lst_output --temp /custom_apps/lst_temp --device cpu
Limiting the number of threads to 12

N=7 NUMBER OF CORRECTIONS=5 INITIAL VALUES F= -10792.4 GNORM= 16.6195

I NFN FUNC GNORM STEPLENGTH
1 2 -10806.997 13.588 0.060
2 3 -10835.964 9.530 1.000
3 4 -10844.645 6.900 1.000
4 5 -10847.318 5.110 1.000
5 6 -10850.005 1.613 1.000
6 7 -10850.256 0.913 1.000
7 8 -10850.518 1.275 1.000
8 9 -10850.829 2.024 1.000
9 10 -10851.002 1.145 1.000
10 11 -10851.134 0.605 1.000
11 13 -10851.147 0.439 0.472
12 14 -10851.149 0.285 1.000
13 17 -10851.165 0.255 2.172
14 18 -10851.172 0.388 1.000
15 20 -10851.180 0.317 0.339
16 21 -10851.184 0.240 1.000
17 23 -10851.184 0.180 0.199
18 25 -10851.184 0.248 0.176
19 28 -10851.185 0.225 0.058
20 29 -10851.185 0.107 1.000
21 30 -10851.185 0.069 1.000
22 32 -10851.185 0.104 0.281
23 34 -10851.185 0.050 0.371
24 35 -10851.185 0.062 1.000
25 38 -10851.185 0.066 0.065
26 41 -10851.185 0.100 0.120
27 42 -10851.185 0.029 1.000
28 43 -10851.185 0.035 1.000
29 45 -10851.185 0.033 0.221
30 46 -10851.185 0.098 1.000
31 47 -10851.185 0.022 1.000
32 48 -10851.185 0.023 1.000
33 50 -10851.185 0.021 0.063
34 54 -10851.185 0.020 0.232
35 56 -10851.185 0.036 0.135
36 58 -10851.185 0.113 0.127
37 59 -10851.185 0.025 1.000
38 60 -10851.185 0.009 1.000
39 61 -10851.185 0.011 1.000
40 63 -10851.185 0.021 0.082
41 64 -10851.185 0.011 1.000
42 65 -10851.185 0.030 1.000
43 67 -10851.185 0.009 5.000
44 69 -10851.185 0.096 0.249
45 70 -10851.185 0.117 1.000
46 71 -10851.185 0.045 1.000
47 74 -10851.185 0.146 21.000
48 77 -10851.185 0.131 0.410
49 83 -10851.185 0.149 0.021
50 87 -10851.185 0.159 0.006
51 88 -10851.185 0.145 1.000
52 92 -10851.185 0.095 0.003
53 93 -10851.185 0.086 1.000
54 95 -10851.185 0.076 0.307
55 97 -10851.185 0.069 0.160
56 98 -10851.185 0.062 1.000
57 100 -10851.185 0.028 0.305
58 101 -10851.185 0.028 1.000
END OF LEVEL 0
Level 0 LastIter Metrics -10851.185220 Energy = -10851.185220
Level 0 Final RAS Transform:
0.9985 0.0521 0.0183 3.4550
-0.0525 0.9984 0.0188 13.2356
-0.0173 -0.0197 0.9997 27.0290
0.0000 0.0000 0.0000 1.0000

N=7 NUMBER OF CORRECTIONS=5 INITIAL VALUES F= -10778.8 GNORM= 5.16266

I NFN FUNC GNORM STEPLENGTH
1 2 -10782.704 2.734 0.194
2 3 -10785.193 1.989 1.000
3 4 -10786.876 2.169 1.000
4 5 -10788.100 1.243 1.000
5 6 -10788.466 0.373 1.000
6 7 -10788.499 0.177 1.000
7 10 -10788.529 0.287 1.958
8 11 -10788.546 0.288 1.000
9 12 -10788.555 0.326 1.000
10 13 -10788.566 0.286 1.000
11 14 -10788.570 0.382 1.000
12 15 -10788.575 0.125 1.000
13 17 -10788.576 0.098 0.333
14 18 -10788.577 0.114 1.000
15 19 -10788.577 0.100 1.000
16 20 -10788.578 0.087 1.000
17 21 -10788.578 0.124 1.000
18 22 -10788.579 0.062 1.000
19 23 -10788.579 0.082 1.000
20 24 -10788.579 0.081 1.000
21 25 -10788.579 0.030 1.000
22 26 -10788.579 0.032 1.000
23 28 -10788.580 0.021 0.074
24 29 -10788.580 0.024 1.000
25 30 -10788.580 0.033 1.000
26 31 -10788.580 0.012 1.000
27 33 -10788.580 0.010 0.227
28 34 -10788.580 0.013 1.000
29 35 -10788.580 0.008 1.000
30 37 -10788.580 0.012 0.291
31 38 -10788.580 0.010 1.000
32 39 -10788.580 0.016 1.000
33 40 -10788.580 0.012 1.000
34 41 -10788.580 0.008 1.000
35 42 -10788.580 0.004 1.000
36 46 -10788.580 0.025 1.602
37 49 -10788.580 0.025 0.023
38 50 -10788.580 0.023 1.000
END OF LEVEL 1
Level 1 LastIter Metrics -10788.579580 Energy = -10788.579580
Level 1 Final RAS Transform:
0.9983 0.0567 0.0123 3.4544
-0.0567 0.9984 -0.0064 13.7152
-0.0127 0.0057 0.9999 22.0855
0.0000 0.0000 0.0000 1.0000

N=7 NUMBER OF CORRECTIONS=5 INITIAL VALUES F= -10701.9 GNORM= 4.15881

I NFN FUNC GNORM STEPLENGTH
1 2 -10705.227 2.916 0.240
2 3 -10711.585 3.350 1.000
3 5 -10718.708 5.865 0.451
4 7 -10723.395 6.930 0.492
5 8 -10731.249 5.865 1.000
6 9 -10737.942 2.002 1.000
7 10 -10738.851 0.667 1.000
8 11 -10739.041 0.700 1.000
END OF LEVEL 2
Level 2 LastIter Metrics -10739.040989 Energy = -10739.040989
Level 2 Final RAS Transform:
0.9988 0.0480 0.0120 3.5851
-0.0479 0.9988 -0.0051 10.4865
-0.0122 0.0045 0.9999 9.2825
0.0000 0.0000 0.0000 1.0000
Limiting the number of threads to 12
vnl_lbfgs: Error. Netlib routine lbfgs failed.
Limiting the number of threads to 12

N=7 NUMBER OF CORRECTIONS=5 INITIAL VALUES F= -10768.3 GNORM= 36.2245

I NFN FUNC GNORM STEPLENGTH
1 3 -10937.526 28.385 0.138
2 5 -11010.808 24.661 0.205
3 6 -11058.090 21.267 1.000
4 7 -11110.197 18.539 1.000
5 8 -11135.360 8.379 1.000
6 9 -11141.413 3.525 1.000
7 10 -11142.074 1.671 1.000
8 11 -11142.136 1.318 1.000
9 12 -11142.209 0.684 1.000
10 14 -11142.217 0.419 0.324
11 15 -11142.219 0.325 1.000
12 16 -11142.221 0.275 1.000
13 17 -11142.223 0.089 1.000
14 18 -11142.223 0.100 1.000
15 19 -11142.223 0.152 1.000
16 20 -11142.223 0.068 1.000
17 24 -11142.224 0.207 1.850
18 25 -11142.224 0.177 1.000
19 26 -11142.225 0.147 1.000
20 28 -11142.225 0.136 0.389
21 29 -11142.225 0.089 1.000
22 31 -11142.226 0.081 0.374
23 33 -11142.226 0.073 0.315
24 34 -11142.226 0.119 1.000
25 35 -11142.226 0.111 1.000
26 38 -11142.226 0.070 0.058
27 39 -11142.226 0.066 1.000
28 40 -11142.226 0.065 1.000
29 41 -11142.226 0.057 1.000
30 42 -11142.226 0.032 1.000
31 43 -11142.226 0.057 1.000
32 44 -11142.226 0.087 1.000
33 46 -11142.226 0.079 0.280
34 47 -11142.226 0.092 1.000
35 49 -11142.226 0.054 0.513
36 50 -11142.226 0.037 1.000
37 52 -11142.226 0.031 0.270
38 54 -11142.226 0.023 0.235
39 55 -11142.226 0.019 1.000
40 56 -11142.226 0.020 1.000
41 60 -11142.226 0.034 0.004
42 63 -11142.226 0.021 1.876
43 64 -11142.226 0.022 1.000
44 67 -11142.226 0.051 0.882
45 69 -11142.226 0.065 0.456
46 70 -11142.226 0.037 1.000
47 74 -11142.226 0.031 0.025
IFLAG= -1 LINE SEARCH FAILED. SEE DOCUMENTATION OF ROUTINE MCSRCH ERROR RETURN OF LINE SEARCH: INFO= 3 POSSIBLE CAUSES: FUNCTION OR GRADIENT ARE INCORRECT OR INCORRECT TOLERANCESEND OF LEVEL 0
Level 0 LastIter Metrics -11142.225666 Energy = -11142.225666
Level 0 Final RAS Transform:
0.9991 0.0402 0.0094 3.6260
-0.0406 0.9978 0.0516 11.1969
-0.0073 -0.0519 0.9986 7.2341
0.0000 0.0000 0.0000 1.0000

N=7 NUMBER OF CORRECTIONS=5 INITIAL VALUES F= -11372.3 GNORM= 15.1961

I NFN FUNC GNORM STEPLENGTH
1 2 -11374.182 21.484 0.066
2 3 -11378.979 8.304 1.000
3 4 -11381.134 7.330 1.000
4 5 -11388.074 11.559 1.000
5 6 -11391.727 9.126 1.000
6 8 -11392.089 7.213 0.213
7 9 -11392.767 1.915 1.000
8 10 -11392.848 0.921 1.000
9 11 -11392.870 0.404 1.000
10 12 -11392.877 0.215 1.000
11 13 -11392.879 0.216 1.000
12 14 -11392.880 0.128 1.000
13 15 -11392.881 0.075 1.000
14 16 -11392.881 0.092 1.000
15 17 -11392.881 0.059 1.000
16 18 -11392.881 0.020 1.000
17 19 -11392.881 0.017 1.000
18 21 -11392.881 0.015 0.329
19 23 -11392.881 0.006 0.440
20 24 -11392.881 0.007 1.000
21 25 -11392.881 0.004 1.000
22 26 -11392.881 0.002 1.000
23 27 -11392.881 0.002 1.000
24 28 -11392.881 0.003 1.000
25 30 -11392.881 0.004 0.317
26 32 -11392.881 0.004 0.428
27 34 -11392.881 0.003 0.368
28 35 -11392.881 0.002 1.000
29 37 -11392.881 0.002 0.236
30 39 -11392.881 0.002 0.491
31 42 -11392.881 0.001 0.042
32 43 -11392.881 0.002 1.000
END OF LEVEL 1
Level 1 LastIter Metrics -11392.881160 Energy = -11392.881160
Level 1 Final RAS Transform:
0.9988 0.0464 0.0127 3.7442
-0.0466 0.9988 0.0134 10.9368
-0.0121 -0.0140 0.9998 8.2613
0.0000 0.0000 0.0000 1.0000

N=7 NUMBER OF CORRECTIONS=5 INITIAL VALUES F= -11463.6 GNORM= 16.9441

I NFN FUNC GNORM STEPLENGTH
1 3 -11465.446 9.463 0.012
2 4 -11466.221 7.975 1.000
3 5 -11468.231 7.389 1.000
4 6 -11470.292 6.646 1.000
5 7 -11472.139 16.246 1.000
6 8 -11473.171 7.127 1.000
7 9 -11473.394 0.847 1.000
8 10 -11473.403 0.607 1.000
9 11 -11473.417 0.541 1.000
END OF LEVEL 2
Level 2 LastIter Metrics -11473.416639 Energy = -11473.416639
Level 2 Final RAS Transform:
0.9989 0.0464 0.0089 3.7775
-0.0465 0.9989 0.0025 10.8482
-0.0088 -0.0029 1.0000 8.1516
0.0000 0.0000 0.0000 1.0000
Limiting the number of threads to 12

########################
If you are using hd-bet, please cite the following paper:
Isensee F, Schell M, Tursunova I, Brugnara G, Bonekamp D, Neuberger U, Wick A, Schlemmer HP, Heiland S, Wick W,Bendszus M, Maier-Hein KH, Kickingereder P. Automated brain extraction of multi-sequence MRI using artificialneural networks. arXiv preprint arXiv:1901.11341, 2019.
########################

File: /custom_apps/lst_temp/sub-X_ses-Y_space-mni_T1w.nii.gz
preprocessing...
image shape after preprocessing: (129, 153, 129)
prediction (CNN id)...
0
1
2
3
4
running postprocessing...
exporting segmentation...
Limiting the number of threads to 12
Limiting the number of threads to 12

@sbajaj1 - great news 🚀 - happy you have obtained your first segmentations using LST-AI. That's also some great feedback for us regarding MAC-OS and we will work on merging this to master in the upcoming days.

Regarding your questions:

(1) Output files:

lst_out/
├── space-orig_desc-annotated_seg-lst.nii.gz --> seg. in native FLAIR space, annotated according to McDonalds criteria
└── space-orig_seg-lst.nii.gz --> binary seg. in native FLAIR space 

The class labels for the annotated segmentation are:

Output labels:
    1  ==  Periventricular
    2  ==  Juxtacortical
    3  ==  Subcortical
    4  ==  Infratentorial

(2) Temp files:

LST.AI performs segmentation and annotation in the MNI152 template space. For users that would like to keep the registered T1w and FLAIR images and their corresponding segmentation masks, we allow to keep these files if the user provides a temp directory. If you skip the --temp option, then LST-AI will create and delete the temp folder for you (and the temp files vanish).

├── affine_flair_to_mni.mat --> Registration Matrix obtained via greedy
├── affine_t1w_to_mni.mat --> Registration Matrix obtained via greedy
├── atlas_mask_warped.nii.gz --> Deformation Field obtained via greedy for annotation of lesions acc. to McD criteria
├── atlas_warp_field.nii.gz --> Deformation Field obtained via greedy for annotation of lesions acc. to McD criteria
├── sub-X_ses-Y_space-mni_brainmask.nii.gz --> Brain Mask in MNI space
├── sub-X_ses-Y_space-mni_desc-stripped_FLAIR.nii.gz --> Skull-stripped FLAIR in MNI
├── sub-X_ses-Y_space-mni_desc-stripped_T1w_mask.nii.gz --> Brainmask T1w in MNI space
├── sub-X_ses-Y_space-mni_desc-stripped_T1w.nii.gz --> Skull-stripped T1w Mask
├── sub-X_ses-Y_space-mni_FLAIR.nii.gz --> FLAIR in MNI space
├── sub-X_ses-Y_space-mni_seg-annotated.nii.gz --> Anno. Seg. in MNI Space
├── sub-X_ses-Y_space-mni_seg.nii.gz --> Binary Segmentation in MNI Space
├── sub-X_ses-Y_space-mni_T1w.nii.gz --> T1w (with skull) in MNI Space
├── sub-X_ses-Y_space-org_FLAIR_mask.nii.gz --> Brainmask in Native Space
├── sub-X_ses-Y_space-org_T1w_mask.nii.gz --> Brainmask T1w in Native Space
├── sub-X_ses-Y_space-orig_desc-stripped_FLAIR.nii.gz --> Stripped FLAIR in native space
├── sub-X_ses-Y_space-orig_desc-stripped_T1w.nii.gz --> Stripped FLAIR in native space
├── sub-X_ses-Y_space-orig_FLAIR.nii.gz --> copy of orig. FLAIR
├── sub-X_ses-Y_space-orig_seg-annotated.nii.gz --> output result: Ann. Seg. in native FLAIR space 
├── sub-X_ses-Y_space-orig_seg.nii.gz --> output result: Binary Seg. in FLAIR space
└── sub-X_ses-Y_space-orig_T1w.nii.gz --> copy of T1w 

(3) You are absolutely correct, LST-AI currently does not gather the statistics as the previous LST/SPM12 version.
However, as you (and likely other users) require this, I quickly implemented this as a separate script here.

To compute your stats, you can run the script using the following command:

Stats for binary mask

python3 compute_stats.py --in space-orig_seg-lst.nii.gz --out test_binary.csv

Example Output:

Stats for multi-class mask

python3 compute_stats.py --in space-orig_desc-annotated_seg-lst.nii.gz --out test_annotated.csv --multi-class

Example Output:

Let me know if this is what you would expect, and if you think that we should also integrate this into the LST-Ai package.

Cheers,
Julian

Hi @jqmcginnis @neuronflow

Thanks for the detailed information.

Somehow, the temp folder I got has 14 files in it (not 20 as you described) as can be seen in my previous screen shot I shared, and 0 files in output folder (not 2 as you described). I also got that error in the terminal console as following: vnl_lbfgs: Error. Netlib routine lbfgs failed.

It seems something minor still needs to be fixed. Please let me know otherwise.

I will next try computing the stats for binary mask/lesion.

Thanks,
Sahil

@sbajaj1 This might be an issue with your input data.

@jqmcginnis @CompImg can we(here meaning you) supply example data to check the functionality? This would allow us to pinpoint whether it is a data issue.

@sbajaj1 can you describe your input data a bit please? resolution etc.

Hi @neuronflow

I am using standard dataset from here: https://www.kaggle.com/datasets/ilknuricke/neurohackinginrimages to test LST-AI This is open access data to public. I double checked and it seems this input data is OK. Please find the data info below:

T1w:
Volume information for BRAINIX_NIFTI_T1.nii.gz
type: nii
dimensions: 512 x 512 x 22
voxel sizes: 0.468750, 0.468750, 6.000001
type: SHORT (4)
fov: 240.000
dof: 1
xstart: -120.0, xend: 120.0
ystart: -120.0, yend: 120.0
zstart: -66.0, zend: 66.0
TR: 450.00 msec, TE: 0.00 msec, TI: 0.00 msec, flip angle: 0.00 degrees
nframes: 1
PhEncDir: UNKNOWN
FieldStrength: 0.000000
ras xform present
xform info: x_r = -0.9997, y_r = -0.0017, z_r = 0.0239, c_r = 2.2905
: x_a = 0.0000, y_a = 0.9974, z_a = 0.0720, c_a = 0.3165
: x_s = 0.0240, y_s = -0.0720, z_s = 0.9971, c_s = 34.8959
Orientation : LAS
Primary Slice Direction: axial

voxel to ras transform:
-0.4686 -0.0008 0.1436 120.8842
0.0000 0.4675 0.4322 -124.1257
0.0112 -0.0338 5.9827 -25.1514
0.0000 0.0000 0.0000 1.0000

voxel-to-ras determinant -1.31836

ras to voxel transform:
-2.1327 -0.0000 0.0512 259.0993
-0.0037 2.1278 -0.1536 260.6955
0.0040 0.0120 0.1662 5.1879
-0.0000 -0.0000 -0.0000 1.0000

FLAIR:

Volume information for BRAINIX_NIFTI_T1.nii.gz
type: nii
dimensions: 512 x 512 x 22
voxel sizes: 0.468750, 0.468750, 6.000001
type: SHORT (4)
fov: 240.000
dof: 1
xstart: -120.0, xend: 120.0
ystart: -120.0, yend: 120.0
zstart: -66.0, zend: 66.0
TR: 450.00 msec, TE: 0.00 msec, TI: 0.00 msec, flip angle: 0.00 degrees
nframes: 1
PhEncDir: UNKNOWN
FieldStrength: 0.000000
ras xform present
xform info: x_r = -0.9997, y_r = -0.0017, z_r = 0.0239, c_r = 2.2905
: x_a = 0.0000, y_a = 0.9974, z_a = 0.0720, c_a = 0.3165
: x_s = 0.0240, y_s = -0.0720, z_s = 0.9971, c_s = 34.8959
Orientation : LAS
Primary Slice Direction: axial

voxel to ras transform:
-0.4686 -0.0008 0.1436 120.8842
0.0000 0.4675 0.4322 -124.1257
0.0112 -0.0338 5.9827 -25.1514
0.0000 0.0000 0.0000 1.0000

voxel-to-ras determinant -1.31836

ras to voxel transform:
-2.1327 -0.0000 0.0512 259.0993
-0.0037 2.1278 -0.1536 260.6955
0.0040 0.0120 0.1662 5.1879
-0.0000 -0.0000 -0.0000 1.0000
MDACM0CL9004107:sample_test sbajaj$ mri_info BRAINIX_NIFTI_FLAIR.nii.gz
Volume information for BRAINIX_NIFTI_FLAIR.nii.gz
type: nii
dimensions: 288 x 288 x 22
voxel sizes: 0.798611, 0.798611, 6.000001
type: SHORT (4)
fov: 230.000
dof: 1
xstart: -115.0, xend: 115.0
ystart: -115.0, yend: 115.0
zstart: -66.0, zend: 66.0
TR: 9500.00 msec, TE: 0.00 msec, TI: 0.00 msec, flip angle: 0.00 degrees
nframes: 1
PhEncDir: UNKNOWN
FieldStrength: 0.000000
ras xform present
xform info: x_r = -0.9997, y_r = -0.0017, z_r = 0.0239, c_r = 2.2900
: x_a = 0.0000, y_a = 0.9974, z_a = 0.0720, c_a = 0.6455
: x_s = 0.0240, y_s = -0.0720, z_s = 0.9971, c_s = 34.8722
Orientation : LAS
Primary Slice Direction: axial

voxel to ras transform:
-0.7984 -0.0014 0.1436 115.8765
0.0000 0.7965 0.4322 -118.8096
0.0192 -0.0575 5.9827 -25.4152
0.0000 0.0000 0.0000 1.0000

voxel-to-ras determinant -3.82668

ras to voxel transform:
-1.2518 -0.0000 0.0300 145.8192
-0.0022 1.2489 -0.0902 146.3430
0.0040 0.0120 0.1662 5.1879
-0.0000 -0.0000 -0.0000 1.0000

@sbajaj1 apologies that I misread and misunderstood your previous comment - somehow I interpreted that it finally worked after I had seen the many temp files! I am sorry about that. Please let me know if I overlooked anything else!

So, I tested a couple of things on my end:

(1) I have used the BRAINIX_NIFTI files from Kaggle - and on my end, they seem to work fine, i.e. I get all temp and output files.

However, I think it might not be the best sample dataset, as it looks to me as if the scan does not feature MS lesions, and also exhibit highly anisotropic data. Perhaps you would like to switch to MSLUB and choose one of the patients under Raw Data? 🙂

(2) According to the developer of greedy, the warning vnl_lbfgs: Error. Netlib routine lbfgs failed. is not necessarily critical:

"This might not be a problem, unless the registration is doing something nonsensical. Registration is an optimization problem, and sometimes the tolerances for the optimization scheme are not quite right for the problem. Here it looks like it ran for some iterations and reduced the objective, so it might be ok. If you see it printing this error after just a couple of iterations (NFN), then something is wrong."

I get the same warning as well. However, the registration (also in your case) seems to work, as the files are populated in your temp folder. I think we can ignore it 👍

(3) Checking the debug output it seems that the script stops after the registration / after warping the images and before entering the LST Segmentation routine ( I am missing the "Running LST Segmentation" print):

File: /custom_apps/lst_temp/sub-X_ses-Y_space-mni_T1w.nii.gz
preprocessing...
image shape after preprocessing: (129, 153, 129)
prediction (CNN id)...
0
1
2
3
4
running postprocessing...
exporting segmentation...
Limiting the number of threads to 12
Limiting the number of threads to 12

Two prints of Limiting the number of threads to 12 indicate that the apply_warp() function has been entered twice, and as it happens twice, I would assume that this function also works on your MacBook.

Next, apply_mask() is called (c.f. here) and after that, and before we enter the actual segmentation function, there should be another print output print("Running LST Segmentation.") c.f. here.

Based on this, my feeling is that the segmentation is still on-going (and the prints are lagging behind).

Could you please provide additional information about the duration of the operation within the Docker container?
Specifically, it would be helpful to know how long you waited for the process to complete.
Additionally, were there any other messages or debug outputs that appeared in the command line interface after some time had passed? It's possible that the Docker container might not have completed its operation yet.

In our experience, especially noted when running Docker on @neuronflow's M2 system, there can be a delay in the appearance of Docker's print commands. Therefore, it's conceivable that the LST-AI process is still ongoing. Could you run it again and wait (for a longer time?).

Thank you very much for your help, and sorry once again for my misunderstanding.

Julian

My Macbook has an M1 Max, otherwise, I agree :)

@sbajaj1, which resources did you assign to Docker on your system?

My successful tests were with these settings, I am quite sure one could get away with much less:

Hi @neuronflow @jqmcginnis

So, finally, it seems everything is working fine now. Two things I noticed:

1. Not every data can be processed using LST-AI (though I am not sure why?) e.g. subject # 1 from this data set here: https://github.com/muschellij2/open_ms_data?tab=readme-ov-file#raw-data did not work but subject # 15 worked fine.
2. As @neuronflow suggested, I changed the memory limit from 12 GB to 20 GB and subject # 15 worked fine - but earlier I had default memory limit set to 12 GB which was not enough and stopped the processing in between. Setting it to 20 GB worked without any issue.

Also, stats command compute_stats.py is working fine.

Thank you so much for your hard working resolving all the issues.

Best,
Sahil

I just downloaded T1(pre) and FLAIR from subject # 1.

During the processing I notice greedy shows the following warnings:

 IFLAG= -1  LINE SEARCH FAILED. SEE DOCUMENTATION OF ROUTINE MCSRCH ERROR RETURN OF LINE SEARCH: INFO= 3 POSSIBLE CAUSES: FUNCTION OR GRADIENT ARE INCORRECT OR INCORRECT TOLERANCESEND OF LEVEL   0
Level   0  LastIter   Metrics  -10694.596561  Energy = -10694.596561

Also I see:

vnl_lbfgs: Error. Netlib routine lbfgs failed.

However, it still generates segmentations that look fine to my layman's eyes:
space-orig_desc-annotated_seg-lst.nii.gz
space-orig_seg-lst.nii.gz

What happens if you increase the resources available to Docker?

Hi @neuronflow @jqmcginnis

Everything is working fine with all the subjects I tested now (there was something wrong at my end while downloading the sample data in pat 1 vs pat 15).

Thank you so much all your consistent help and support with this.

Best,
Sahil

sweet, thank you for reporting back so quickly :)

@jqmcginnis I believe this issue can be closed?

@neuronflow I am currently working on a PR for all changes related to this issue in order to push it to the main branch of this repo. For referencing the PR and problems addressed in this issue, I would keep it open until I created & closed the PR? :)

@neuronflow @twiltgen

The improvements implemented in this issue have been merged to the main branch, and have now been released in v1.1.0 🎈