Pre-operative segmentation not detecting manually-set sequence type (T1-CE) about raidionics

[andreped]

@ft-zhang Not sure what the issue is, but @dbouget should be able to assist you.

I observed that anaconda seems to be used instead of the Python interpreter bundled with Raidionics. That was surprising to me. Have you seen something like this before @dbouget? Windows is kind of strange with its Path variable.

Maybe there is something wrong with our PyInstaller setup? I have previously seen that if two Python versions are on the same build machine, both are bundled with. Maybe not related but PyInstaller Spec file might have a solution for this.

---

EDIT: It might be that the local Python is used when you spin a new subprocess! That would made a lot of sense as this is outside PyInstaller's control. After Installation, is the Python version inside the raidionics.exe file or outside of it? If outside, I think we have a solution for this :]

[ft-zhang]

Sorry -- forgot to add that I am not using the Windows executable but a conda environment running main.py. Thanks again for your help.

[andreped]

Sorry -- forgot to add that I am not using the Windows executable but a conda environment running main.py. Thanks again for your help.

Which main.py? Are you trying to run Raidionics (the actual GUI software) by downloading the installer, installing it, and running the app? Or do you mean that you are trying to run Raidionics from source, by first installing all dependencies and then running python main.py? Basically what is described here:
https://github.com/raidionics/Raidionics/wiki/Developers'-corner#version-10--

If you try to run Raidionics from source, you have to remember that Raidionics only works for a very specific configuration! If you want to do this, you need to use Python 3.8 (and 3.8 only), you have to use the basic Python and not Python from Anaconda. You also have to follow these installation steps as seen here (or at least you should to ensure that everything works fine):
https://github.com/raidionics/Raidionics/blob/master/.github/workflows/build_windows.yml#L33

For any other use, please download the installer of Raidionics for Windows from here:
https://github.com/raidionics/Raidionics/releases

And then run the program as you would with any other software. No need to run from python directly, unless you have very specific requirements on what you are trying to do. If so, I would likely to know more what you are trying to do, because it might be that Raidionics already supports this :]

[ft-zhang]

Thank you @andreped for your continued attention and support.

I am indeed running from source by installing the Raidionics dependencies manually then running python main.py. I am attempting this as a troubleshooting measure as the issue in my original post above is also occurring with Raidionics installed from the stable Windows installer (i.e. uploading a patient MRI, changing the sequence type to T1-CE, running pre-operative segmentation, and failing silently). The Windows CLI window associated with Raidionics outputs no error messages, hence I am attempting to debug from source.

More broadly -- I would like to be able to programmatically deploy Raidionics segmentation/other tools as part of a pipeline/workflow, and not have to use the GUI. Guidance on this front would also be appreciated.

Thank you again!

[dbouget]

Hi @ft-zhang,

Running Raidionics without GUI

If you are willing to integrate the processing from Raidionics without the need for the GUI, then I recommend you directly look into this repository.
The Raidionics RADS library contains the processing backend for running the different segmentation and reporting tasks, while Raidionics itself "simply" bundles the backend with a GUI in addition.

You will find several notebook tutorials to get started which are also available on Colab. Here is the example for running a preoperative segmentation model, on Github and Colab

Debugging the missing radiological volume error

Since you will be running the Python code, you can set break points along the way. I would recommend to set one here, so that you can investigate the content (i.e., MRI volumes and annotations) associated with the current patient to be processed.

If you want to investigate how a folder is parsed to populate a PatientStructure, which could explain why the T1-CE volume is not properly found or associated as T1-CE, you can set a break point here.

Let me know if it helped identifying the issue!